Sunday, August 05, 2007

AMD: Limited Options

AMD trailed Intel's 65nm process by ten months in 2006 but the recent speed of conversion at FAB 36 has been impressive. Likewise, the announcement of only 2.0Ghz for Barcelona's release this quarter was disappointing. However, Fuad's article shows normal improvement for Barcelona with each stepping. This is good news because moving forward is really AMD's only option.

The microprocessor volume shares from the last seven years are interesting:

2000 - AMD 16.5%, Intel 83.5%
2001 - AMD 20.7%, Intel 79.3%
2002 - AMD 16.4%, Intel 83.6%
2003 - AMD 17.0%, Intel 83.0%
2004 - AMD 16.1%, Intel 83.9%
2005 - AMD 18.5%, Intel 81.5%
2006 - AMD 23.4%, Intel 76.6%
2007 - AMD 21.0%, Intel 79.0%

We can see that up to 2004, AMD averaged about 16.5% to Intel's 83.5%. AMD had a slight bump in 2001 during K7's competitive period against PIII and Williamette P4 but then dropped back down again in 2002 against Northwood P4. Some people mistakenly think that AMD's fortunes improved right after it released K8 in 2003. This was clearly not the case since AMD's volume share was lower in 2004 than it was in 2003. In 2004 Intel was on its 90nm process and gaining volume from 300mm FABs like FAB 11X and FAB 24 plus D1C which had been freed up for production by the new development FAB, D1D. AMD, in contrast, was expanding FAB 30 as fast as it could in the new clean room space and trying not to lose volume with a K8 die that was twice the size of the K7 die. This did not really come together for AMD until 2005 when the late 2004 move to AMD's own 90nm process allowed a much smaller die and the new expansion ultimately allowed a 50% increase in capacity. Ho weer, the 2006 numbers are misleading because the volume fell off from 2005. Without this dropoff, AMD's volume share would have been only around 19-20% since FAB 30 was topped out and the new production from FAB 36 did not arrive until the second half of 2006.

Some will no doubt see the latest numbers as just another bump like 2001. They will most likely assume that AMD will drop back down to its previous volume of 16%. However, that is not possible. There is a scene in the movie Vertical Limit where a woman has to jump from the tiny ledge she is standing on out into empty air and jam the crampon that is attached to her harness in a crack in the cliff face before she falls to her death. This is pretty much what AMD did when it built FAB 36 and purchased ATI. But, AMD had good reason for such a radical move. When AMD first introduced K7 and K8, Intel used its considerable muscle to prevent 3rd party companies from supporting them. Thus, no motherboard makers attended K8's launch and even companies like Acer were told not to attend. In both cases, AMD had to supply its own chipsets to get these platforms off the ground and there were still delays and areas not covered. For example, although AMD's K6 chip was capable of dual socket operation, there was no supporting chipset that allowed it. Likewise, AMD's 760 MP chipset was the only dual socket chipset produced for K7. Considering that K7's were used to build supercomputers, it would be difficult to suggest that the lack of other MP chipsets was due to K7 itself. Essentially, AMD's Dresden Design Center was able to produce a bare minimum chipset support for AMD's processors. However, this has inadequate compared to the much greater support that Intel provided to its processors. AMD only got out of this box by purchasing ATI. That AMD gained a much more robust capability to develop its own supporting infrastructure can be seen with the 690G, RD790, and 780M chipsets. AMD's choice was to either buy ATI and gain competitive support for its processors or die a slow death.

FAB 36 was also not much of a choice. Processor volume grows slowly most years at 3-11%. AMD did a very good job in increasing FAB 30's capacity by 50%. However expansion alone could not erase Intel's 300mm wafer advantage or prevent FAB 30 from eventually topping out. There was no doubt that AMD badly needed 300mm wafer facilities to remain competitive. It is theoretically possible that AMD could have added 300mm tooling to FAB 30 during its expansion in 2003 and 2004 and this would have reduced costs. However, FAB 30 would still have topped out and AMD would again be faced with the prospect of slipping every year in volume share until it became marginalized. For AMD, to not gain the necessary processor support nor to increase capacity meant marginalization and the eventual loss of any capability to compete in front line processors. AMD's ability to fund research and development would have slipped until AMD was like VIA is today or like AMD used to be before K6.

Unfortunately, neither of these could be done in stages or phased in. You can't really build 10% of a FAB nor can you buy 10% of a graphics company. This is why AMD had to take a leap out into empty air and hope it didn't plummet to the rocks below. I've seen some people try to lay the blame for AMD's financial troubles entirely on the ATI purchase. Seemingly, these people do not realize that AMD will spend $1.8 Billion in this year alone for tooling to outfit FAB 36; this is in addition to what it spent in 2006 and will spend in 2007 plus what the FAB itself originally cost. You can reasonably look at the issue either way. AMD could have easily afforded ATI if it hadn't built FAB 36 or AMD could easily have afforded FAB 36 if it hadn't bought ATI. Having to pay for both is what is difficult. I've wondered many times if there was any other option that could have gotten AMD out of the trap it was in. However, delaying the ATI purchase would have meant delayed benefit as well and AMD desperately needs the 780 mobile chipset that will come in 2008. Similarly, Fusion will not pay off until 2009 but, with Intel's moving in the same direction, pushing this back any further would have meant even greater competitive disadvantage to Intel. So, having built FAB 36 and bought ATI, AMD can now no longer survive on 16% of the market. AMD must move forward with at least 25% volume share or else it has no future.

That AMD intends to gain volume share is without question since in spite of the losses in Q4, Q1, and Q2 AMD's production plan is the same as it was in October 2005. AMD still plans to make 100 Million chips in 2008. This is more than double the number of chips that AMD made in 2005. However, the market is not growing that fast. The only way that AMD can sell 100 Million chips in 2008 is if it takes share from Intel. Naturally, Intel would prefer otherwise and will do its best to prevent this from happening. Looking at the volume share, one might easily think that nothing has changed over many years. However, this impression would be wrong. There has indeed been a strong dynamic element to the market even though Intel has garnered an advantage at each step. For example, Intel originally benefited by being ahead in terms of processor performance. This lasted until AMD introduced K6 in 1997. However, by that time Intel was making Pentium Pro and chipsets so it gained income from server and chipset sales that AMD couldn't match. It would be another four years before AMD released Athlon MP in 2001 and had a server processor offering of its own. However, with only a single chipset for support, Athlon MP never gained much marketshare and it would be another three years until AMD actually began taking server share in 2004 with K8.

By 2002 Intel had a steadily increasing advantage from 300mm wafer manufacturing compared to AMD's 200mm facility. AMD is only now benefiting from a majority of 300mm wafers in production. Likewise, it has only recently purchased ATI to gain stronger chipset support. Intel still has some advantages in mobile but these will essentially be gone in 2008 with AMD's new mobile platform. AMD will finally release a small volume of K10 server chips this quarter. In Q4, AMD should have more chipset offerings plus more K10 chips. However, Intel will begin releasing 45nm chips in Q4 and should hold onto the highest clock speed. It gets more interesting in Q1 08 because Intel will have a good volume of 45nm chips by then while AMD should have both faster K10 chips and a reasonable number of desktop K10 chips. On the face of it Intel could lose its speed advantages in Q1 08 but with 45nm ramping it should experience lower costs. AMD is unlikely to match 45nm until Q3 08. Intel's entire strategy seems to be based on lower cost 45nm chips plus low prices to put pressure on AMD. This was certainly effective up to now but may be less so as AMD's chip ratios move more strongly towards 65nm and 300mm wafers. Presumably, Intel swaps its 65nm to 90nm advantage for a 45nm to 65nm advantage. However, Intel does lose the 300mm advantage. This is a gain for AMD but then AMD hasn't made any profit for three quarters. The final factor though is that K10 should have more value versus Penryn than K8 did against Conroe.

With AMD's recent $1.5 Billion losses the notion of any gain for AMD seems counterintuitive. Yet if we list Intel's advantages:

Intel Chipsets – Declining as AMD moves forward with ATI
Intel mobile – Less than before, gone in 2008
300mm wafers – Declining rapidly, gone by 2008
65nm – Currently declining but will begin increasing again with 45nm in Q4
Quad core – Gone this quarter
C2D higher value – Declining once K10 is released, gone by Q2 08

So, except for the 45nm process, Intel is losing most of its advantages. This should make AMD much stronger by the start of 2008. Intel does still have Nehalem which could be a great processor and put Intel back in the lead. However, it looks like AMD may be better prepared this time with Shanghai than it was with K8. On the face of it I can't see any reason why AMD would not be able to gain share. And, since I've already shown that AMD has no middle ground it is a certainty that they will be striving their best to do just that. I think we will also find out conclusively in Q4 whether Intel has truly cut costs and leaned out the company. I think the last thing that Intel wants is to have a bad 4th quarter and they will have to have excellent cost control to maintain a price war with AMD in Q4. If AMD's only options are to gain share or go out of business we also have to look at who would lose if AMD failed. It's pretty obvious that neither IBM, Sun, Cray, HP, Dell, Gateway, nor Apple would be happy with the absence of AMD. Other than Intel, the only companies I can think of that would benefit might be VIA and possibly Silicon Graphics. With that number of companies benefiting from AMD the idea that it would go out of business soon does seem unrealistic.

190 comments:

enumae said...

Scientia & Abinstein

Before I make a post, I am curious if either of you could look at this table and see if you find a problem. I don't want to strat off on the wrong foot :)

Thanks.

abinstein said...

enumae -
"I am curious if either of you could look at this table and see if you find a problem."

For a starter I don't think you can linearly scale up SPECcpu score with respect to clock frequency. Please take a look at this article that I wrote 3 months ago. For dual-socket dual-core Core 2, 1.5x clock frequency results in 1.3x SPECint and just 1.22x SPECfp. Note that for quad-core, the scaling will be worse. This lack of perfect scaling is expected when the gap between core and memory speed becomes larger, i.e., relative memory access latency increases with higher core clock frequency.

The implication is that higher IPC will be preferable to higher clock frequency. Note also that K8 scales somewhat better than Core 2 in both SPECint and SPECfp (1.4x and 1.35x for 56% higher clock frequency). Assume 2.6GHz K10 has SPECint ~101, a 1.9GHz K10 is likely to have SPECint ~87 (rather than just 72.7). Similarly, assume 2.66GHz Penryn has SPECint ~110, a 3.16GHz Penryn will be around 122.

Secondly, I'm not sure you can assume Penryn that much (10%) higher IPC for SPECint. I believe the extra 2MB cache is not to improve performance but (together with the faster FSB) to allow scaling performance above quad-core 2.66GHz. The SSE4 instructions will have negligible effects for SPECcpu and most programs, whether one likes it or not.

Also, looking at Xeon 5355 in April and Xeon 5355 in July, it's obvious that the 5-10% higher scores of the latter come from optimization of Intel C++ compiler v10 toward the SPECcpu. Just because icc optimizes for SPEC better doesn't mean your binary will run faster - it won't.

Ahmar Abbasi said...

http://www.hkepc.com/bbs/hwdb.php?tid=837360&tp=Intel-Penryn-Preview&rid=837360

Penryn preview.......bring out those Spec simulated vaporware AMD CPUs with AMD approved slideshow with absolutely nothing concrete for comparison.....

I wonder if AMD will release a counter processor....i mean slideshow....to combat this new development..........

abinstein said...

It seems the only reason that one would upgrade from Conroe to Penryn is to encode from mpeg2 to mpeg4 with SSE4-enabled encoders.

If you really want fast encoding. however, buy a cheap K8 plus an AMD graphics card, and use CTM instead. You'll get something like 5-10x the speedup.

All other "improvements" of Penryn are around 10% or less, most not even 5%. This is pathetic. Lets hope for Intel's sake Penryn is as cheap as E6550, or there are enough fools to buy into this "45nm" marketing.

As for games, there is simply no noticeable difference between 132fps and 181fps. Again, buy a cheap K8 and an AMD graphics card (or nvidia's), and you'll get better performance on higher video quality.

Ho Ho said...

abinstein
"The SSE4 instructions will have negligible effects for SPECcpu and most programs, whether one likes it or not"

THat will depend on how well can compiler use SSE4 instructions in the code. ICC has done quite well with autovectorization thus far.


"Just because icc optimizes for SPEC better doesn't mean your binary will run faster - it won't."

It can. I know a few people who have reported similar gains with ICC10 with their SSE2 heavy code. You know Spec stuff isn't the only thing that gets improved with new compiler versions.

Ho Ho said...

abinstein
"It seems the only reason that one would upgrade from Conroe to Penryn is to encode from mpeg2 to mpeg4 with SSE4-enabled encoders."

Ok, so we have one killer application for Penryn. Is there any for K8/10?


"If you really want fast encoding. however, buy a cheap K8 plus an AMD graphics card, and use CTM instead. You'll get something like 5-10x the speedup."

Is there anything usable on the market already?


"All other "improvements" of Penryn are around 10% or less, most not even 5%."

Average comes to around 9.8%, 7.4% without DivX. Not too bad for a "small core update with only more cache and faster FSB", as some have said earlier. What did you expect?

Also how much clock to clock improvement do you expect K10 to have over K8/Core2/Penryn? More than that 7.4(9.8)% on average over those benchmarks? Will it be big enough not to be able to say "this is pathetic" for a huge core upgrade and (claimed) IPC improvement? How much speed increase should have Penryn had for you not to call the improvement pathetic?

Another thing you seemingly missed is power usage. Performance per watt has increased significantly with Penryn, much more than raw performance. I wouldn't be surprised to see AMD having trouble getting close to that level in the near future with K10, especially considering that ES chips have usually had much higher power usage than retail versions.


"As for games, there is simply no noticeable difference between 132fps and 181fps."

It just shows that Penryn increases HL2 performance a lot. It is a physics intensive game with lots of FP calculations, you can bet that cache is not the only reason for huge boost. If a game comes out on Source engine that is four times as recource hungry as that HL2 demo then Penryn will run it around 45FPS vs 33 on Core2. K8 will be significantly slower.

Scientia from AMDZone said...

ringinheat

"bring out those Spec simulated vaporware AMD CPUs with AMD approved slideshow with absolutely nothing concrete for comparison"

I'm sorry but you are incorrect. The latest AMD SPEC numbers were taken directly from AMD K10 demo systems just as Intel did with V8.

Unknown said...

Intel should commence a hostile takeover of AMD. Buy 50% of the shares by offering the shaderholders a 50% premium over AMD's current share price.

Once that's done they can terminate AMD's x86 license as per the terms of the agreement. Then AMD is finished.

Scientia from AMDZone said...

giant

"Intel should commence a hostile takeover of AMD. Buy 50% of the shares by offering the shaderholders a 50% premium over AMD's current share price."

I'm sure Intel would do that if it were possible (even if it weren't legal). Unfortunately, such a purchase would have to be approved by the SEC and it would not be. Intel would also wouldn't mind buying nVidia to gain an immediate boost in discrete but unfortunately Intel's share of the graphics market is too large; that wouldn't be approved either.

"Once that's done they can terminate AMD's x86 license as per the terms of the agreement. Then AMD is finished."

Yes, Intel already tried that. I'm sure Intel would prefer to have no competitors.

Ahmar Abbasi said...

All other "improvements" of Penryn are around 10% or less, most not even 5%. This is pathetic. Lets hope for Intel's sake Penryn is as cheap as E6550, or there are enough fools to buy into this "45nm" marketing.

This statement says it all........Intel never once claimed more than 10% improvement but yet when AMD claims performance leads over intel's highest parts and after much delay is only able to muster a 2.0ghz part to compete against Intel's lower mid range of processors that isnt some how pathetic in your world.......

Unknown said...

10% performance improvement from Penryn is most impressive. It's not a massive boost, it doesn't need to be. Intel already has a commanding lead.

Re: Nvidia, I'm sure Intel could acquire Nvidia if they truly wanted to, and Nvidia agreed to be acquired. Intel's graphics business is only in cheap lowend IGPs. But it would be a massive cost. Nvidia is worth over $15bn, and adding in a reasonable premium you get a price of about $20bn for Nvidia. Intel could afford that, but it would be a huge drain on their financial resources resources.

There's just no reason for Intel to do that. They have a fine relationship with Nvidia as it is.

Aguia said...

Penryn performance boost looks interesting, power consuming even more.
Games performance looks very good; just don’t understand why with games the boost was so high. Because it there where huge SSE improvements the Sisoft tests should had show them.

Ho Ho said...

scientia
"I'm sure Intel would prefer to have no competitors"

There is only one reason for all the companies existens: to make their owners rich. Everything else they say is pure marketing to sell more products, earn more and thus get richer. Without competitors that would be much easier and I can assure you AMD would also like Intel to dissapear so the owners would get richer easier and faster.

Roborat, Ph.D said...

Abistein said: "It seems the only reason that one would upgrade from Conroe to Penryn is to encode from mpeg2 to mpeg4 with SSE4-enabled encoders....

you are being foolish to assume that all potential Penryn buyers are Conroe owners. Most buyers with very old systems will see a massive jump in performance.

All other "improvements" of Penryn are around 10% or less, most not even 5%. This is pathetic"

You're standards for Intel are pretty high considering there are uarch tweaks involved. Compared to AMD who saw a performance drop when they did a dumb shrink on an old Athlon layout. It would be interesting to find out what your baseline is when you said "pathetic". Otherwise it just sounds like "sour grapes" to me.

Christian H. said...

One thing I notice you didn't harp was that according to Mercury Research AMD got back 4.5% market share for Q207.

Now there is a 3.2GHz Opteron (reviewed at Anand) that remains at 120W and a new 3GHz that runs at 95W.

The power efficiency is incredible and goes to what I mentioned about AMD wanting to keep those upgraders.

I also see that even with a broadcast of SPEC running, that still doesn't prove anything.

AMD has balls, plain and simple but they also have the support of OEMs.

I never doubted that 10h would be 30-70% faster per core - depending on the application.

Hanging in this long with Intel shows they know what they're doing. They hire well. (The CEO doesn't need to design chips) Doug Grose, Dirk Meyer are the key players for AMD.

They seem to be delivering what they need to. More efficient CPUs at lower prices. That's what will keep their share.

They have already stated that Barcelona would debut at the original Opteron prices (that old Hector Ruiz video).

They can do this IF perf is higher per clock. Especially if they can maintain their idle power advantage with Barc.

enumae said...

Abinstein
All other "improvements" of Penryn are around 10% or less, most not even 5%. This is pathetic.

Did AMD gain between 5 and 10% going from 90nm to 65nm?

Going from DDR1 to DDR2?

-----------------------------------

Thanks for looking at the table. I will work to revise it according to your explanations.

Aguia said...

Did AMD gain between 5 and 10% going from 90nm to 65nm?

Going from DDR1 to DDR2?


While your point is quite correct enumae, there is also one major factor for that:
AMD didn’t do nothing.

Also Intel didn’t improve with DDR2.
DDR3 I still have doubts too.
The chipset improved with DDR3/DDR2.

Ho Ho said...

christian m. howell
"They can do this IF perf is higher per clock. Especially if they can maintain their idle power advantage with Barc."

Does anyone know if the new low power state in Penryns will be laptop-only feature or will it be as with AMD where all the laptop powerstates are also availiable on the recent desktop CPUs?


aguia
"Also Intel didn’t improve with DDR2.
DDR3 I still have doubts too. "


Lower power usage.


Another thing to consider with the "pathetic" 10% speed increase is that it took Intel around three such pathetic increases to get from Pentium3 to K8 killer.

abinstein said...

roborat -
"Compared to AMD who saw a performance drop when they did a dumb shrink on an old Athlon layout."

I thought many Intel fanboys claimed that Penryn was much more than a dumb shrink, and it would offer up to 20% performance boost, and even outperform K10 clock-for-clock??? Now the truth is out, precisely as I've predicted, that Penryn is a Core 2 with larger cache and minor tweaks.

Out of all tests, only two other than SSE4-supported encoding have improvement around 10%. All others are strictly less than 10%, half of which are even less than 5%. The 4MB larger cache alone can probably contribute the difference.

AMD made a dumb shrink on the initially 130nm core design to 65nm. The shrink was not meant to bring performance, as I have said way before AMD's 65nm transition. I know there have been AMD fans wishing the other way, and later found themselves dead wrong. Their errors doesn't make those of Intel fanboys less, however.

Unknown said...


I thought many Intel fanboys claimed that Penryn was much more than a dumb shrink, and it would offer up to 20% performance boost


Provide proof of this. Fact: Penryn delivers, on average, a 10% IPC boost across the board. Fact: When coupled with a 10% increase in clock speed (Rumored 3.33Ghz Yorkfield Extreme Edition quad in Q4'07) you get a CPU about 20% faster, on average, versus the fastest desktop CPU now; the QX6850. Obviously this figure could be higher or lower depending on the application in question.

This does compare, in stark contrast, to AMD's 65nm K8 shrinks that are SLOWER than the 90nm versions! The clockspeeds are lower and they use slower L2 cache.

AMD themselves admitted that K10 will only offer a 15% IPC advantage over K8.

The proof of that is right here, in AMD's own power point presentation:

http://se.sun.com/virtualisering/pdf/AMD_Quad_Core-Leif_Nordlund.pdf

Look at slide 16.

"Projected IPC gains up to 15%"

Obviously this is a per core IPC gain.

Given that Core 2 currently enjoys a 20% IPC advantage of K8, Penryn increases that by another 10% and we have Penryn with a 15% IPC advantage over K10. In other words, to match desktop 3.33Ghz Yorkfield AMD needs a 3.83Ghz Phenom to match it.

Axel said...

abinstein

I thought many Intel fanboys claimed that Penryn was much more than a dumb shrink, and it would offer up to 20% performance boost, and even outperform K10 clock-for-clock???

And there were AMD fanboys who claimed that the AM2 refresh of K8 would offer a major IPC improvement over 939 and keep up with C2D. Like that fruitcake known as The Ghost on your favorite site AMDZone. He's still eating crow to this day for trumpeting those false claims; I hope he learned his lesson that what's demoed in the labs isn't necessarily what will be practical to fabricate.

So what's your point? Fanboys can claim anything. Intel themselves have, IIRC, only claimed a minor improvement in IPC except when SSE4 is utilized, in which case the IPC gain is truly massive, a generational leap. In fact, I would class the performance improvement in HL2 alone to be a generational leap despite it not using SSE4. I would guess here that the new Super Shuffle Engine is helping tremendously with SSE2 and SSE3 throughput for games like HL2.

Aguia said...

Does anyone have a link to a review where they test AMD and/or Intel performance in Virtualization?

Mo said...

Very well said Axel. It's saddens me to see people hang on to words of a Fanboy and use that as an example of what Intel or AMD said.

Just read this
http://www.tcmagazine.com/comments.php?shownews=15647&catid=2
Seems that nehalem should tap out soon or has already tapped out.

Whats the word on Shanghai?

abinstein said...

Axel -
"And there were AMD fanboys who claimed that the AM2 refresh of K8 would offer a major IPC improvement over 939 and keep up with C2D."

I guess you are just incapable of reading. Now see what I said, copied below for your convenience, in the very same comment that you tried to respond to:

AMD made a dumb shrink on the initially 130nm core design to 65nm. The shrink was not meant to bring performance, as I have said way before AMD's 65nm transition. I know there have been AMD fans wishing the other way, and later found themselves dead wrong.

abinstein said...

giant -
"Fact: Penryn delivers, on average, a 10% IPC boost across the board."

What? Is it 10% across the board, or more accurate, mostly around 5% and rarely 10% (2 out of 20-some cases)?

It's clear that you are just a sad fanboy in denial. Penryn is precisely as what I have predicted, a (very) minor tweak of Core 2 with larger cache and SSE4.

Axel -
"I would class the performance improvement in HL2 alone to be a generational leap despite it not using SSE4."

I have to say anyone who believes "Penryn improves gaming" according to the link is dumb, because going from 120fps to 180fps is a waste, period. Test Penryn with a good graphics card on high resolution (at least 1280x1024) and highest video quality, and see how much the min frame rates improves. If it improves from 30-something to 40-something, then it improves gaming. (I don't know, maybe it does, but the point is we don't know yet, and anyone who thinks otherwise is nothing but dumb fanboy).

Christian H. said...

giant

Given that Core 2 currently enjoys a 20% IPC advantage of K8, Penryn increases that by another 10% and we have Penryn with a 15% IPC advantage over K10. In other words, to match desktop 3.33Ghz Yorkfield AMD needs a 3.83Ghz Phenom to match it.




You're looking at it from a K8 POV not a K10 POV. Phenom is faster than K8 per core so you can't base it on K8.

Ho Ho said...

abinstein
"Penryn is precisely as what I have predicted, a (very) minor tweak of Core 2 with larger cache and SSE4"

Seeing as you are so good at predicting, how much of and IPC increase against Core2/K8 will we see with K10? Will it be more than pathetic 5-10%? At what increase it won't be pathetic any more? Could one say that for a whole new core you'd need to increase IPC a lot more than for a refresh for it not to be pathetic?


"I have to say anyone who believes "Penryn improves gaming" according to the link is dumb, because going from 120fps to 180fps is a waste, period."

Waste or not but still a huge increase in performance.


"I don't know, maybe it does, but the point is we don't know yet, and anyone who thinks otherwise is nothing but dumb fanboy"

Either a fanboy or just making predictions, similar to what you are doing by calling it a pathetic CPU based on the same data as they are.


You still haven't answered any of the questions in my first posts.

Axel said...

abinstein

I guess you are just incapable of reading...

AMD made a dumb shrink on the initially 130nm core design to 65nm. The shrink was not meant to bring performance, as I have said way before AMD's 65nm transition.


And you are forgetting some CPU basics. AM2 had nothing to do with 65nm. AM2 was the summer 2006 refresh of K8 that introduced a new IMC with support for DDR2. According to a legion of AMD fanboys led by The Ghost who had apparently witnessed impressive performance gains in the AMD labs, AM2 was also supposed to bring a whole host of other improvements to the core to improve IPC.

Early previews of AM2 by Anandtech and others showed minimal to no improvements to IPC, and these previews were proved 100% correct upon release.

To Scientia's credit, he didn't buy in too much to The Ghost's misguided exuberance.

Now with regard to Brisbane 65nm, that was just a big disappointment. A new 65nm K8 die shrink with a lower IPC and less clocking headroom than the 90nm predecessor? It saved a bit of power but performed worse per clock! That along with the ludicrously inefficient and power-hungry 4x4 concept made December 2006 a complete washout for AMD from a PR standpoint.

abinstein said...

Ho Ho -
"Either a fanboy or just making predictions, similar to what you are doing by calling it a pathetic CPU based on the same data as they are."

Whether you like it or not, an average of 5% to 7% performance improvement is pathetic. This is not prediction, but some actual data.

"Waste or not but still a huge increase in performance."

I believe that means you are willing to pay a premium for the additional frame rates beyond 120fps. It's fine with me, though IMO you are just dumb to do so. (Of course if Penryn is as cheap as E6850 then it's not a bad thing to have this uselessly higher fps.)

"You still haven't answered any of the questions in my first posts."

Did you actually notice that this is my first comment to reply to you? Maybe you haven't noticed but your credit has been squandered with me, and I don't feel much desire to read less respond to your "questions" at all. I have in the past pointed out so many errors of yours, which are nothing but half-baked thoughts and FUDs. I'm tired of them.

abinstein said...

axel -
"A new 65nm K8 die shrink with a lower IPC and less clocking headroom than the 90nm predecessor?"

65nm K8 has slightly less IPC because it employs an L2 cache with higher latency, and added circuitry for better power management and deeper sleep states.

K8 was not an ideal design for 65nm anyway. It was initially designed for 130nm, then scaled down quite successfully to 90nm. K10 OTOH is designed for 65nm/45nm, and you can expect 45nm Shanghai to have roughly the same improvement over Barcelona, as Penryn over Conroe. (In other words, minor tweaks).

Ho Ho said...

abinstein
"Whether you like it or not, an average of 5% to 7% performance improvement is pathetic"

Compared to what? When putting it against K8 from the day it was born to the latest ones with numerous core updates and refreshes that 5-7% is rather impressive I'd say.


"I believe that means you are willing to pay a premium for the additional frame rates beyond 120fps. It's fine with me, though IMO you are just dumb to do so."

I don't get it. Are you just playing dumb or is there something else wrong?

As I said earlier if there would be a Source engine based game taxing CPU four times as much that FPS increase will be significant. Also HL2 likely won't be the only game that would see similar performance increase. There are many more games already out there and more to come that rely heavily on physics and they will also see quite a bit of increase in performance. HL2 is just one of them.


"Did you actually notice that this is my first comment to reply to you?"

Probably another fruit of my tired mind but perhaps you are simply avoiding uncomfortable questions?


"I have in the past pointed out so many errors of yours, which are nothing but half-baked thoughts and FUDs. I'm tired of them."

Then why not stop making all sorts of questionable statements yourself? Also why not simply use PAPI to measure real world applications in real world situations instead of coming up with stuff like you have in your latest blog entry. All you do is quote one book and take it as gospel.

Scientia from AMDZone said...

roborat

"Compared to AMD who saw a performance drop when they did a dumb shrink on an old Athlon layout."

Perhaps. The first 65nm testing did show a difference of less than 3% due to changes in L2 cache access. This by definition means that it was not a dumb shrink. However, the 65nm transistors have also improved since December but there has been no new testing. The difference in speed may or may not still be present.

Are Penryn's gains good?

Well, fairly good. K8 only got a boost of about 4% with Revision F in 2006 so 8% for Penryn would be twice as much.

The overall increase of IPC for K10 over K8 is about the same as Penryn over Yonah. We could easily see speed differences between K10 and Penryn at the same clock of 3% or less (which is below the accuracy of the benchmarks).

AMD does have Shanghai and claims that this too increases IPC but it isn't clear yet if this is a general increase or is only in certain areas. Then of course we have Nehalem and Bulldozer which have many similarities like 8 core support, IMC, point to point bus, support for GPU's, etc.

abinstein said...

Ho Ho -
"As I said earlier if there would be a Source engine based game taxing CPU four times as much that FPS increase will be significant."

And do you know if there is one?
Just claiming there could be one doesn't mean it is useful. If there are games that tax the cpu in the way fluid dynamics does then K8 will perform better than Core 2. How is that?

I reckon you're a graphics programmer. So why don't you work on something that really benefits Penryn in terms of gaming? Make something that runs satisfactory on Penryn (e.g., 40fps or higher) but not on K8 or Core 2 (e.g., 30 fps and lower). And if your games is an interesting/fun one, then I'll believe you that Penryn does improve gaming.


"Probably another fruit of my tired mind but perhaps you are simply avoiding uncomfortable questions?"

You know what, my AI toy program probably can do better than you in terms of intelligent conversation.

The reason that I did not respond to your questions was stated right in the statement you quoted yourself: I'm tired of all your half-baked thoughts and FUDs.


"Then why not stop making all sorts of questionable statements yourself?"

What? Penryn's 5% and 7% except SSE? Or Clovertown's poor scalability on both # cores and clock frequency? Or the fact that you were dead wrong to say processor performance is most limited by memory bandwidth?


"Also why not simply use PAPI to measure real world applications in real world situations instead of coming up with stuff like you have in your latest blog entry. All you do is quote one book and take it as gospel."

Maybe you can send a letter to professor Hennessy and ask him to measure real world application performance instead of making questionable statements? Or maybe you can quote some other reputable books saying the reverse (of that CPUs benefit from lower latency more than higher bandwidth) is true? Or maybe you have some results "simply using" PAPI that shows DDR2-800 bandwidth is insufficient?

Wake me up when you have some evidence to prove yourself correct.

Scientia from AMDZone said...

And, let's stop the "dumb" and "fanboy" comments. This is not going to turn into a flamefest.

Abinstein, I don't understand why you are trying to argue that gains in video speed are meaningless. While it is true that these don't matter to most computer buyers they are not useless for everyone.

You only have to note the triple video card Crossfire demo by AMD which has vastly more video power than most people need. On the other hand, could this level of video performance push into the traditional graphic workstation market? Would this amount of power be useful for heavy CAD or rendering preview or video editing?

A game demo can be an indicator of performance in other areas even if it might not be useful on that particular game. It can also be an indicator of future game performance.

Axel said...

Scientia

I'm sorry but you are incorrect. The latest AMD SPEC numbers were taken directly from AMD K10 demo systems just as Intel did with V8.

No, AMD has updated the Analyst Day slide deck.
http://download.amd.com/Investor%20Relations/July_2007_AMD_Analyst_Day_Randy_Allen_FINAL.pdf

See slide 20 for the latest SPECfp_rate2006 benchmark and note the new footnote: Estimated performance based on AMD lab measurements of K10. In other words, they apparently didn't actually run SPECfp in the lab but either:
- Simulated it using K8;
- Ran other software on K10 and inferred the improvement to SPECfp.

Scientia from AMDZone said...

giant

"AMD themselves admitted that K10 will only offer a 15% IPC advantage over K8."

I don't believe you can make this statement without some sleight of hand.

"The proof of that is right here, in AMD's own power point presentation:

http://se.sun.com/virtualisering/pdf/AMD_Quad_Core-Leif_Nordlund.pdf"


Yes, this presentation does state 15% IPC in the specific context of servers. Your point is valid if you are only talking about server chips.

"Obviously this is a per core IPC gain."

Yes, quite true.

"Given that Core 2 currently enjoys a 20% IPC advantage of K8,"

Oops, here comes the sleight of hand. You attempt to turn a server chip comparison into a general comparison.

Conroe does enjoy a 20% IPC advantage over X2 Athlon 64. However, Woodcrest does not enjoy a 20% advantage over X2 Opteron. Opteron has about the same IPC as Woodcrest.

"Penryn increases that by another 10% and we have Penryn with a 15% IPC advantage over K10.

No, we have Wolfdale Xeon with a 10% advantage over K8 Opteron, and K10 Opteron with a 15% advantage over Woodcrest. This gives a K10 Opteron with a 4.5% advantage over Wolfdale Xeon.

"In other words, to match desktop 3.33Ghz Yorkfield"

then the 15% is irrelevant.

abinstein said...

axel -
"See slide 20 for the latest SPECfp_rate2006 benchmark and note the new footnote: Estimated performance based on AMD lab measurements of K10."

No, you are wrong. The 69.5 SPECfp is the real measurement, as confirmed by AMD. The "estimated" was added there because the benchmark scores are not submitted and not verified by SPEC.

Axel said...

Abinstein
No, you are wrong. The 69.5 SPECfp is the real measurement, as confirmed by AMD. The "estimated" was added there because the benchmark scores are not submitted and not verified by SPEC.

After GutterRat's post I went to AMDZone and found that thread. You can believe whatever you wish, but I would take that kind of second-hand news with a grain of salt.

Until we have an official announcement (or Powerpoint slide lol) from AMD, this info from someone at AMD is about as credible as Fudzilla in my book. Did the AMD rep who spoke with Polonium210 have the proper authority and knowledge to make such a statement? The Ghost's claims of AM2 performance gains were also based on information from within AMD itself, but were proven dead wrong.

Scientia from AMDZone said...

axel

I thought I deleted your post by mistake but then I realized after I reposted it from my copy that I didn't. That is good because I didn't intend to delete it.

You are quite right. AMD did indeed change the slide after the presentation. I had to edit my original Analyst Day article because of that. However, AMD then provided an update to the SPEC scores from a live demo.

GutterRat

No apologies to you. Behave and stop trolling or don't post. And, if you can find something on AMDZone (or anywhere else) that disproves what I said above feel free to post it.

Scientia from AMDZone said...

You can see the live SPEC DEMO yourself at vnunet. CPU face off: AMD Opterons vs. Intel Clovertown. Again, this is just like the live V8 demo that everyone referenced.

abinstein said...

"I don't understand why you are trying to argue that gains in video speed are meaningless. While it is true that these don't matter to most computer buyers they are not useless for everyone."

I didn't say video encoding is meaningless. I said the increase of gaming frame rate from 120fps to 180fps is useless.

The video encoding speedup is cool, and it is as I said one reason to buy Penryn is one wants to encode his MPEG4 faster but don't want to use more effective hardwares such as AMD's CTM.

abinstein said...

Giant -
"Given that Core 2 currently enjoys a 20% IPC advantage of K8, Penryn increases that by another 10% and we have Penryn with a 15% IPC advantage over K10."

Remember what you said here, because I'm quite sure you'll find yourself dead wrong when Phenom and Penryn are released.

According to motherboard makers who have tested both chips, Phenom does have somewhat better IPC than Penryn. The latter's advantage is SSE4 (which Phenom lacks) and higher clock frequency. If however AMD can speed up the release Phenom at 3.0GHz, Penryn will lose the lead, which may not be taken back until Nehalem.

lex said...

LOL you pretty much summarized why AMD will lose.

Being late by 6-9 months to a technology node is bad enough. Your die are twice the size and yield will suffer too. That is a huge penality to make up. Add that to the fact you have such a smaller volume to distribute that higher cost and try and eck out a profit and you are guranteed loss.

Also when your process at any given process node is inferior then you are double penalized.

You paper monkeys have no clue to silicon. Back in 130nm or earlier days you could buy a process from Nikon, AMAT etc. These days HighK / Metal gate will take amd till 32nm to figure out. At 45nm they are late by 3 quarters at the very least and handicapped by 20% performance.

Sorry AMD design can't beat INTEL design by that much.

Too much advantage for INTEL, AMD will be selling processors for far cheaper then they orginally planned and will have to further curtail R&D and manufacturing capacity expenditures. That is a downward spiral that will cause them to fall further and further behind.

AMD is finished...

Scientia from AMDZone said...

lex

"Your die are twice the size"

No. Yorkfield needs 214 sq mm for quad core while Barcelona needs 291 sq mm. I get 36% more die area for Barcelona.

"and yield will suffer too."

Barcelona yields are excellent. I assume Intel is doing okay with 45nm as well.

"Add that to the fact you have such a smaller volume to distribute that higher cost"

??? I don't know what you mean. This will probably be about 50% of production in Q2 08. That's about 10 Million chips.

"Also when your process at any given process node is inferior then you are double penalized."

Intel claims 20% more speed with 45nm high K and metal gate.

AMD claims 20% more speed with 45nm without high K or metal gate.

However, Intel should have lower power consumption. This will probably be enough for Intel to match AMD's power consumption.

"These days HighK / Metal gate will take amd till 32nm to figure out."

AMD can do high K and metal gate in 2009 if Intel is much ahead. Otherwise, AMD will wait until 32nm.

"At 45nm they are late by 3 quarters at the very least and handicapped by 20% performance."

Well, not late because Q3 is AMD's original plan. However, it would be about three quarters after Intel. And, you are incorrect about the 20%

"Too much advantage for INTEL,"

26% less die area versus AMD's 65nm chips. Presumably AMD's Shanghai will have about the same die area. And, you get a bit lower power draw due to the process which is good because Intel needs it.

Nehalem will have more advanced power conservation however it will also be a monolithic die.

I suppose someone could say that Intel is at least 3 quarters behind AMD in terms of monlithic quad core. However, we don't know yet if this will actually be an advantage. It does appear that K10 is at least 3 quarters ahead of Intel in advanced power conservation. So, this does neutralize some of Intel's 45nm process advantage.

" AMD will be selling processors for far cheaper then they orginally planned"

Why would they be? At this point Intel has only shown a 10% increase in clock speed with Penryn. Unless Intel will be able to release chips faster than 3.33Ghz in Q1 I can't see why AMD would have a problem with pricing.

"and will have to further curtail R&D"

That's another place you are wrong; AMD hasn't curtailed any R&D.

"and manufacturing capacity expenditures."

This is quite true; AMD has pushed back the upgrade schedule for FAB 38 a bit. However, how did you miss that I stated that AMD had not changed its 2008 production estimate from October 2005? It's still 100 Million chips.

" That is a downward spiral"

You certainly haven't shown that.

Unknown said...

No. Yorkfield needs 214 sq mm for quad core while Barcelona needs 291 sq mm. I get 36% more die area for Barcelona.

Yorkfield is two 107mm dies, not a single 214mm die. The advantage of the MCM is the two smaller dies. It's much easier to produce two smaller dies than a single larger die.

You also have the advantage that both dual and quad core CPUs can be produced using the same dies. For AMD to do that would be quiet expensive, unless they have lots of Barcelona CPUs with two defective CPU cores.

AndyW35 said...

I still wonder why the cache latency went up with 65nm versions of K8, the reason given by AMD at the time was slightly strange and I've not seen any more on this since, so it remains a bit of a puzzle.

As for the Penryn scores, it's been common knowledge that on the tock of the 2 year cycle ( or is it tick ? ) Intel would be just reducing the process and tweaking and 10% has been common knowledge for a long time, so I don't think that anyone should be claiming to have had their predictions confirmed at this late stage ( unless it is to people living in a cave ).

The Wofldale power requirements are the biggest news but I would like confirmation from other sources first. The other big issue is how far Intel can go with making them cheaply.

Now that AMD's speed issues seem to be solved and I think most people would widely agree it will be good with power usage the main questions will not be on the performance I think but yield and cost between the two combatants lastest generation of chips.

It's going to be an interesting 6 months. I think a lot of AMD socket 939 and Intel Netburst users will be waiting to see which way to jump when they finally upgrade.

Axel said...

Scientia

Unless Intel will be able to release chips faster than 3.33Ghz in Q1 I can't see why AMD would have a problem with pricing.

Remember that with Intel's new pricing as of July 22, the fastest 1333 FSB dual core and entry level quadcore are both $266 each for trays of 1000 (though currently the actual street prices are significantly higher due to great demand). One more price cut on the Q6600 puts quadcore firmly into mainstream OEM territory.

By the time Q1 2008 rolls around and Phenom is selling in volume (we hope), Intel will likely have already launched Penryn and Kentsfield prices will already have been cut once again. Q6600 and all Conroe dual cores will probably be under $200. It is likely that the 3.33 GHz Wolfdale dual cores will not exceed $266, as Intel evidently intends to exert pricing pressure to the extreme to ensure that AMD cannot make a profit on their Kuma dual cores. Kuma will have to be priced lower than Wolfdale at the same clock and here's why:

Kuma may have similar IPC to Wolfdale for non-SSE apps but will almost certainly be slower clock-for-clock in SSE due to Penryn's new Super Shuffle Engine that accelerates SSE2/SSE3 throughput, as can be seen by the gaming improvements in the HKEPC preview linked earlier by 13ringinheat. For apps re-compiled with SSE4, Penryn will simply destroy K10. From what we have seen so far, Wolfdale simply brings more value to the table than Kuma unless K10 has some hidden rabbit in the hat we don't know about.

Since AMD are unlikely to have Kuma in volume at >3 GHz until Q2 08, they are likely to be stuck with <$150 pricing for all volume dual cores for the foreseeable future. We know they cannot make a profit on that with their current expenditures.

AMD's only hope to bring their overall ASP up is with Agena (desktop quadcore K10). I'm guessing that the bulk of these 283 mm2 monsters will have to be sold at $300 to $500 for AMD to have a prayer of making a profit. Unfortunately, I don't think Intel will allow this to happen. As Penryn is likely to be as fast as Agena clock-for-clock in non-SSE, faster with SSE2/SSE3, and easily dominating with apps re-compiled for SSE4 (e.g. DivX), I believe AMD will have to price Agena lower than Penryn at the same clock. Intel will use their 45-nm die size advantage to their fullest to price AMD down to ASPs that cannot support their current CAPEX / R&D spend profile.

Speaking of die size: Most of AMD's CPUs produced this past quarter have been Brisbane (126 mm2 die) though a substantial percentage were Windsor (183 mm2). AMD lost $600 MM with an overall ASP of $62 on a mix of CPUs that was primarily Brisbane. What does their overall ASP need to be to have a prayer of breaking even on a 283 mm2 K10 die with an additional metal layer, let alone making a profit?

With Fab 38 pushed out to 2009, AMD are maxed out on 65nm capacity for the foreseeable future. In addition, some line capacity in Fab 36 has to be reserved for those huge K10 dies and also 45nm re-tooling. Therefore AMD will not be realizing much more in the way of cost savings with 300-mm / 65nm as Fab36 is already fully converted.

No, I don't see how AMD can make money anytime in the next three quarters at least. I see bankruptcy as a very real possibility. At their Analyst Day the execs were completely forward-looking for 2009 onwards, almost ignoring the present difficulties for 2007 and 2008. This means they are likely desperately trying to find someone to keep them alive financially in the face of credit rating cuts and continuing massive losses, because they have products in the pipeline that can bring them back in the running if only someone lends them the cash they need to bring these plans to fruition.

Scientia from AMDZone said...

giant

"Yorkfield is two 107mm dies, not a single 214mm die."

Correct. However, the number of dies does not change the amount of wafer area required. Intel is still 26% less, not 50% less.

"It's much easier to produce two smaller dies than a single larger die."

Theoretically you could get a better yield with Yorkfield. However, this is a long, long way from Roborat's absurd 3X more expensive estimate. The worst case I can come up with is that Barcelona would cost 63% more than Yorkfield. However, it is unlikely that Intel's yields will be that good in Q4. AMD basically has a headstart in the 65nm process but I would expect Intel to have better yields on 45nm in Q1 and Q2.

In the second half of the year it gets more complicated because the Shanghai die gets samller while Nehalem gets much larger.

"You also have the advantage that both dual and quad core CPUs can be produced using the same dies."

Quite true. However, it doesn't cost AMD anything more (beyond the cost of the masks) to run multiple dies at the same time.

Scientia from AMDZone said...

axel

"Remember that with Intel's new pricing as of July 22"

I already went over this in the comments in the last article. The prices you want are the Q4 45nm prices:

X5460 3.16 GHz, 12 MB, 1333 FSB, $1,164
E5450 3.0 GHz, 12 MB, 1333 FSB, $852
E5440 2.83 GHz, 12 MB, 1333 FSB, $695
E5430 2.66 GHz, 12 MB, 1333 FSB, $467
E5420 2.5 GHz, 12 MB, 1333 FSB, $332
E5410 2.33 GHz, 12 MB, 1333 FSB, $273
E5405 NA, 12 MB cache, NA, $229

"By the time Q1 2008 rolls around and Phenom is selling in volume (we hope), Intel will likely have already launched Penryn and Kentsfield prices will already have been cut once again."

You have a false argument. Essentially you are trying to argue that Intel is insulated from pricing effects. This is not the case. We can see definite indications of pricing effects on Intel in Q2 and this will get worse over the next 3 quarters. Intel will not be able to just painlessly cut prices.

"Kuma may have similar IPC to Wolfdale for non-SSE apps but will almost certainly be slower clock-for-clock in SSE due to Penryn's new Super Shuffle Engine"

Which Penryn needs because it is slower in conversions. K10 can run just as fast with different code.

"Since AMD are unlikely to have Kuma in volume at >3 GHz until Q2 08, they are likely to be stuck with <$150 pricing for all volume dual cores"

I have no idea what you are talking about here. I assume you are claiming that Intel's 3.0Ghz chips will be $150 with only the 3.33Ghz speeds at, say, $200. I suppose that is possible if Intel plans to scrap the Conroe L's and replace them with real dual cores. But, if 3.0Ghz starts at $150 you wouldn't have room to go down to the current 1.6Ghz chips and you would have to drop at least the two lowest speed grades. We haven't seen any indication of that yet but it is possible.

"I'm guessing that the bulk of these 283 mm2 monsters will have to be sold at $300 to $500 for AMD to have a prayer of making a profit."

So, you are assuming that Intel will have to have $184 - $306 to make a proft with Yorkfield? The lowest price in the list above is $229 so I suppose that is possible.

"Intel will use their 45-nm die size advantage to their fullest to price AMD down to ASPs that cannot support their current CAPEX / R&D spend profile."

Intel can't do that without cutting its own profits.

"Speaking of die size: Most of AMD's CPUs produced this past quarter have been Brisbane (126 mm2 die) though a substantial percentage were Windsor (183 mm2). AMD lost $600 MM with an overall ASP of $62 on a mix of CPUs that was primarily Brisbane. "

No. The majority were still 90nm. They will be a majority 65nm this quarter.

"What does their overall ASP need to be to have a prayer of breaking even on a 283 mm2 K10 die with an additional metal layer, let alone making a profit?"

By Q4 AMD needs about $100 ASP to break even.

"With Fab 38 pushed out to 2009, AMD are maxed out on 65nm capacity for the foreseeable future."

?? Where did you get this idea? FAB 36 will increase in capacity until mid 2008. Secondly, FAB 38 is not pushed out to 2009. It will begin producing chips in Q3 08 which is two quarters later than originally planned.

" In addition, some line capacity in Fab 36 has to be reserved for those huge K10 dies and also 45nm re-tooling."

This is already figured in. You seem be trying to count the K10 die size two or three times.

"Therefore AMD will not be realizing much more in the way of cost savings with 300-mm / 65nm as Fab36 is already fully converted."

No. The tooling already in FAB 36 is 65nm however it is still ramping and will for the next year.

"No, I don't see how AMD can make money anytime in the next three quarters at least."

AMD should turn a profit in Q1 08.

"At their Analyst Day the execs were completely forward-looking for 2009 onwards, almost ignoring the present difficulties for 2007 and 2008."

You must have looked at a different Analyst Day. Most of the material covered the second half of this year with a lesser amount covering 2008 and 2009.

"This means they are likely desperately trying to find someone to keep them alive financially"

Not yet. They'll be fine if they break even in Q4.

"because they have products in the pipeline that can bring them back in the running if only someone lends them the cash they need to bring these plans to fruition."

Again, where did you get this idea? Everything should be fine up to Griffin, 780M, 45nm and Shanghai. I suppose AMD is still working on Bulldozer, Bobcat and 32nm.

Scientia from AMDZone said...

andy

"I still wonder why the cache latency went up with 65nm versions of K8"

AMD changed the L2 controller so that they could add more L2 cache. Some have suggested that this was a test bed for K10 since there is no indication that AMD is actually increasing the cache size on Brisbane.

Mo said...

?? Where did you get this idea? FAB 36 will increase in capacity until mid 2008. Secondly, FAB 38 is not pushed out to 2009. It will begin producing chips in Q3 08 which is two quarters later than originally planned.

Sci, Can you provide a link for Q308? Because the articles talk about it being pushed back till 09. Are we making predictions we can't back up?

Mo said...

http://money.cnn.com/news/newsfeeds/articles/newstex/AFX-0013-18466974.htm

Thought you might need a 2nd link on the FAB 38 delay.

Axel said...

Scientia
The prices you want are the Q4 45nm prices

First, Fudzilla isn't a credible source; Fuad has a quite dismal record for veracity. For every rumor he gets right there are two more that he flubs. His many incorrect predictions on K10 availability alone were pretty laughable.

Though those prices look plausible, I have yet to see a list from a reasonably credible source like Digitimes or Dailytech. If you know of one, please link it.

Second, the prices you listed are for Xeons. These are generally higher than the desktop variants.

You have a false argument. Essentially you are trying to argue that Intel is insulated from pricing effects.

No, that's not what I'm saying. Intel have clearly demonstrated their willingness to sacrifice some profitability (and near-term share price upside) in order to regain market share and to force AMD into a stranglehold. The strategy appears to be succeeding. Intel management apparently believe that it is in the best interests of their shareholders to continue this strategy for another few quarters until AMD is crippled and back down to <15% overall unit share. Then Intel can comfortably raise prices and improve profitability.

In addition, Intel are to some degree insulated from pricing effects. Starting in Q1 2008 Intel can to some extent dictate Yorkfield / Wolfdale pricing and use Kentsfield / Conroe to squeeze AMD's pricing. This is because Penryn will likely outrun K10 per clock on SSE2/SSE3 and many multimedia apps will be re-compiled for SSE4 within a year. Integer performance is likely to be similar between the two. K10 only has the advantage on x87 and on apps with very high memory bandwidth requirements. In other words, due to a combination of superiority in SSE performance, parity in integer performance, and higher clocks, Penryn is likely to easily retain the performance crown in mobile, desktop, and workstation. Intel should be able to dictate pricing to a large extent.

I have no idea what you are talking about here. I assume you are claiming that Intel's 3.0Ghz chips will be $150 with only the 3.33Ghz speeds at, say, $200.

No. I explained why AMD will not be able to price Kuma at parity with Wolfdale at the same clock. A 3.0 GHz Kuma is likely to be priced the same as a 2.66 GHz Wolfdale. So if Intel price a 3.0 GHz Wolfdale at $215, AMD would conceivably be forced to price a 3.0 GHz Kuma at $180 or thereabouts. 3.0 GHz Kuma will likely be a low volume part in Q1 08, so therefore AMD's high volume Kumas will all be priced below $150. Assuming, of course, that Intel price Wolfdale that low. But I believe that is Intel's strategy, to push quadcore mainstream and suffocate AMD on capacity by forcing them to produce more Agenas to compete. This of course entails that Intel eat some potential profits in dual core. But their current pricing already clearly demonstrates their willingness to do so. What else can be concluded when a year ago the 2933/1066 MHz Conroe was $1000 and today a 3000/1333 MHz Conroe is $266?

I suppose that is possible if Intel plans to scrap the Conroe L's and replace them with real dual cores.

Intel have a couple options here. They could scrap the 1MB L2 E2000s and move the 2MB L2 E4000s down one price deck. They could simply start discontinuing lower end 4MB L2 Conroes. I think your guess is correct and the recent introduction of the 1333 FSB models hints that this is what will probably happen. There are only three models: 2.33, 2.66, 3.00 GHz. The 2.33 GHz is $163 and the 3.00 GHz is $266. The lack of slower 1333 FSB models suggests that the 2.33 GHz will be the new low end E6000 series CPU. My guess is that after the launch of Penryn, for dual cores we will see:
- E2000 series scrapped or dropped to Celeron pricing
- E4000 series cut to below $100
- E6000 series priced in the $133 to $183 range, with 2.33 GHz being the low and 3.00 GHz being the high.
- Wolfdales priced from $166 for low end to $266 for the 3.33 GHz model.

So, you are assuming that Intel will have to have $184 - $306 to make a proft with Yorkfield?

If Penryn is indeed generally superior to K10 as I believe, Intel will price the Yorkfield range from ~$200 to $1000.

Intel can't do that without cutting its own profits.

Their strategy is to regain market share. They have been cutting profits for over a year now.

The reason their guidance for Q3 and Q4 gross margins is up despite lower Conroe / Kentsfield pricing is because the investments in 45-nm tooling have now been completed for two of the four 45-nm fabs.

No. The majority were still 90nm. They will be a majority 65nm this quarter.

I can believe that but please provide a link to this.

By Q4 AMD needs about $100 ASP to break even.

I find this difficult to believe. Please provide a link.

FAB 36 will increase in capacity until mid 2008.

Does this account for re-tooling for 45-nm? Please provide a link.

AMD should turn a profit in Q1 08.

Wishful thinking. You have not provided a logical argument for how this can occur.

Again, where did you get this idea? Everything should be fine up to Griffin, 780M, 45nm and Shanghai. I suppose AMD is still working on Bulldozer, Bobcat and 32nm.

Time will tell. But you've been saying "everything should be fine" for over a year now on this blog as AMD cheerfully proceeded to lose $2 billion over that same time span. So you will excuse me if I remain skeptical.

abinstein said...

axel: "What does their overall ASP need to be to have a prayer of breaking even on a 283 mm2 K10 die with an additional metal layer, let alone making a profit?"

scientia: "By Q4 AMD needs about $100 ASP to break even."


This is assuming they keep losing money on the Graphics and consumer business.

axel: "First, Fudzilla isn't a credible source; Fuad has a quite dismal record for veracity."

Yeah, right, and overclocker is your credible source. Man, you're even lousy as a FUDer. :p

axel: "This is because Penryn will likely outrun K10 per clock on SSE2/SSE3 and many multimedia apps will be re-compiled for SSE4 within a year."

What are you talking about? Where did you see Penryn outruns K10 on SSE2/3? And how many applications do you think can benefit from SSE4?

As I said, Penryn seems only attractive if you want to transcode mpeg2 to mpeg4, but do not wish to use a more effective solution such as AMD's CTM. As indicated by motherboard makers who tested both chips, in general-purpose computing, K10 outperforms Penryn clock-for-clock, even though slightly.

axel: "If Penryn is indeed generally superior to K10 as I believe, Intel will price the Yorkfield range from ~$200 to $1000."

So initially Penryn has SSE4 which K10 doesn't, then it becomes Penryn outruns K10 on SSE2/SSE3, now it turns into "Penryn is indeed generally superior...". Just where is range of your FUDs, kindly let us know at once? Do you have shame at all?

abinstein said...

axel: "But you've been saying "everything should be fine" for over a year now on this blog as AMD cheerfully proceeded to lose $2 billion over that same time span."

Just because scientia made a few wrong predictions doesn't make your FUDs correct. Just wait until you are proven wrong about Barcelona SPEC scores, together with other things (too many FUDs, mostly not worth my memory).

So what's your prediction? That AMD is not going to make a 45nm processor because it is not "competitive" to Intel anyway, even though it is making better graphics at the moment, released the more popular DTX, and occupying high-end servers? Not to mention K10 will simply defeat Xeon on both high thoughput and low power, and Phenom (as claimed by motherboard makers) will help AMD get even on desktop?

"I think your guess is correct and the recent introduction of the 1333 FSB models hints that this is what will probably happen. ... The lack of slower 1333 FSB models suggests that the 2.33 GHz will be the new low end E6000 series CPU."

So you do think that it's possible and plausible that Intel go all the way from 800 to 1333 FSB? Does that make all those "new" Core 2 Duo people bought earlier obsolete less than a year later?

axel: "By Q4 AMD needs about $100 ASP to break even.

I find this difficult to believe. Please provide a link."


I don't understand people like you who 1) demands link but does not think, 2) pick the links he likes to believe. $100 ASP is 60% more than its current $62. This means under the same unit shipment AMD's computing revenue can rise from $1B to $1.6B. Cost of sales will rise, but so will the unit volume, and revenue from other segments like graphics.

Well I know this is too much for you, and I'm sure you will FUD a few more paragraphs away from any good accurate estimate. Remember what you say and you will be found wrong. Remember Glen Yeung who predicted AMD will lose $2.59 per share and said Barcelona is going to be delayed until October? He certainly deserves to be shamed for FUDing, and he is very quiet now.

I expect you will just be like him.

Scientia from AMDZone said...

mo

Slide 153 of AMD's presentation on Analysts Day says:

FAB 38 transformation

Adds flexible capacity to FAB 36 during 2008 and 1H 2009.


The delayed until 2009 story seems to have only one source: Thomson Financial delivered by Newstex. I tried to do a google search on {AMD, FAB 38, 2009, Deppe} and that is the only source I can find.

Single point sources are much more likely to be wrong. If this were really some kind of general announcement there should be other sources.

Scientia from AMDZone said...

axel

"The strategy appears to be succeeding."

Well, yes and no. It has reduced AMD's profitability but it has reduced Intel's as well. Intel's gross margin peaked in Q1 while their ASP peaked in Q4. Both are down in Q2.

"Intel management apparently believe that it is in the best interests of their shareholders to continue this strategy for another few quarters"

I don't think Intel can continue that. AMD's profitability rose in Q2 in spite of lower ASP. Intel has nothing left to increase profitability until it releases 45nm in volume and that is Q1 08.

"until AMD is crippled and back down to <15% overall unit share."

That's impossible. AMD is on a 300mm FAB now; you can't put the genie back in the bottle.

"In addition, Intel are to some degree insulated from pricing effects. Starting in Q1 2008 Intel can to some extent dictate Yorkfield / Wolfdale pricing and use Kentsfield / Conroe to squeeze AMD's pricing."

Q1 08 would be the earliest that Intel would get any gain. Until then, anymore cuts will hurt.

"This is because Penryn will likely outrun K10 per clock on SSE2/SSE3 and many multimedia apps"

Not without SSE4. Recompiling within a year is fine but that also puts it within the time span of Shanghai and Nehalem which is a different fight.

"Intel should be able to dictate pricing to a large extent."

Not with Penryn.

"No. I explained why AMD will not be able to price Kuma at parity with Wolfdale at the same clock."

Right but that is clearly wrong.


"A 3.0 GHz Kuma is likely to be priced the same as a 2.66 GHz Wolfdale."

No. A 2.5Ghz Kuma is likely to be priced the same as a 2.66Ghz Wolfdale.

"What else can be concluded when a year ago the 2933/1066 MHz Conroe was $1000 and today a 3000/1333 MHz Conroe is $266?"

I assume you should conclude that Intel has introduced faster quad cores and these command the top pricing.

"If Penryn is indeed generally superior to K10 as I believe, Intel will price the Yorkfield range from ~$200 to $1000."

My estimate has nothing to do with superiority. It's based solely on die size and worst case yield.

"Their strategy is to regain market share."

Their stategy isn't working.

" They have been cutting profits for over a year now."

Not true. Intel only dropped again in Q2 07.

"The reason their guidance for Q3 and Q4 gross margins is up despite lower Conroe / Kentsfield pricing is because the investments in 45-nm tooling have now been completed for two of the four 45-nm fabs."

The only FAB that will produce chips for Q4 is D1D.

"I can believe that but please provide a link to this."

AMD only got FAB 36 converted to 65nm late in Q2. Those chips don't show up until Q3.

"I find this difficult to believe. Please provide a link."

I can't link to my spreadsheet. The estimate is based on current losses, volume, and ASP. I'm not sure I understand your scepticism with $100 because this would be a 50% increase over what it is now.

"Does this account for re-tooling for 45-nm? Please provide a link."

The retooling is for 300mm but the tools will be compatible with 45nm. However, FAB 38 will lag behind FAB 36. I doubt it will produce 45nm in volume until 2009. Presumably the chips would still be 65nm during 2008.

"Wishful thinking. You have not provided a logical argument for how this can occur."

Slow reduction in expenses, slow increase in gross margin, and slow increase in volume share.

"Time will tell. But you've been saying "everything should be fine" for over a year now on this blog as AMD cheerfully proceeded to lose $2 billion over that same time span."

FAB 36 was delayed in conversion to 65nm - not foreseeable.

Barcelona has required 3 steppings - not foreseeable.

R600 was delayed and didn't perform well on 80hs transistors - not foreseeable.

I was indeed wrong because I could not have predicted any of these things. However, to predict from this point you have to look at the present and it is completely different.

FAB 36 converted in record time with excellent yields.

Barcelona has improved with each stepping and looks good for the B2.

R600 moving to 65nm and the 790 chipset due for Q4.

AMD says 45nm is on track but the proof of that is that Shanghai has taped out. No announcement yet. Also no announcement on Griffin.

Axel said...

abinstein

Only a few of your comments are intelligible enough to respond to:

This is assuming they keep losing money on the Graphics and consumer business.

And assuming they don't lose more market share. High ASPs don't pay the bills when market share has tanked.

Yeah, right, and overclocker is your credible source. Man, you're even lousy as a FUDer. :p

You'd have a point if Ed had made up the ASP figures. But if you'd bothered to click the link you'd have found that Ed had linked to this page containing data from Mercury Research.

Where did you see Penryn outruns K10 on SSE2/3?

According to the HKEPC page, Penryn apparently beats 1333 FSB Conroe by 31% per clock at moderate resolution on Half-Life 2, a game that doesn't use SSE4. There are only three possible explanations for what's supplying the bulk of this improvement:
- The 50% larger L2 cache. Impossible since HL2 is too large a program to be so sensitive to cache. Only SuperPi can benefit so much from L2. Looking at the other benchmarks, we can reasonably discard this explanation by inspection.
- The new Radix 16 divider. No. I don't believe that x87 and integer divide operations are used much in modern game code as SSE and/or multiply operations are generally more effective.
- The Super Shuffle Engine for SSE shuffle operations. By deduction, this is probably the culprit. HL2 apparently contains a great deal of SSE shuffle code.

Now then, Conroe is generally at least some 40% faster than K8 per clock at SSE/2/3 (check out any benchmarks on media encoding out there). With Penryn improving by as much as another 30% without even using SSE4 is a leap that I just don't see K10 making over K8. I could of course be wrong but at this point I think you're the one who's wrong if you believe K10 will catch Penryn in SSE. I doubt even Scientia will support you there. Maybe The Ghost will.

As indicated by motherboard makers who tested both chips, in general-purpose computing, K10 outperforms Penryn clock-for-clock, even though slightly.

Lol should I even bother asking for a link here or just take your word for it. Anyway, your conjecture could actually be right for integer code but I think not for SSE.

Does that make all those "new" Core 2 Duo people bought earlier obsolete less than a year later?

Intel have already made the 1066 FSB Conroes obsolete. The 3.00 GHz E6850 bulk price is $266, while the 2.66 GHz E6700 is $316. They're cannibalizing their own older models in order to put the screws on AMD and regain share, plain and simple.

$100 ASP is 60% more than its current $62. This means under the same unit shipment AMD's computing revenue can rise from $1B to $1.6B.

Congratulations, your first reasonable argument of the day. Ok, I can see the logic in that. Now let's see if AMD can pull it off.

enumae said...

Scientia
AMD only got FAB 36 converted to 65nm late in Q2. Those chips don't show up until Q3.

So, AMD lied in their Q1 earnings call... "and by the way, 100% of our fab 36 wafer starts are on 65 nanometer technology today", that would have been about 19 days into Q2.

Did I misunderstand your post?

Scientia from AMDZone said...

abinstein

Yeung's estimate of a loss of a $2.59 loss was only slightly worse than the consenus estimate of a loss of $2.48. Of course, both of these were wrong since the actual loss was $1.09. And, he was just one month more pessimistic than the September actual availability.

Patrick Wang said August at the earliest which would be correct but then he blows his credibility by saying that Penryn will launch in late third quarter or early fourth quarter. At this point I figure that Intel will try to beat Phenom but neither company has set a specfic date.

Scientia from AMDZone said...

enumae

Yes, my mistake. Early Q2, rather than late Q2.

But, the point is still the same, no more than a few of those chips could show up in Q2 so it wouldn't change the ratio to a majority of 65nm. The first majority 65nm will be Q3. However, the ratio changes rapidly after that as FAB 30 draws down.

Scientia from AMDZone said...

axel

There is no reason for Penryn to gain 31% in SSE; it's SSE pipeline is nearly identical. This might happen with SSE4 in some cases but not in general cases.

Again, the supershuffle helps Intel a lot; AMD doesn't need it. Properly compiled K10 code should be as fast (at the same clock) as Penryn accept for special cases where SSE4 helps. Obviously, ICC would not be proper code (K8 currently takes a 20% hit). We'll have to see if Intel continues its arm twisting to cheat with its compiler or whether someone actually grows a backbone and tests with PGI.

I think your line of reasoning about costs is incorrect however. you seem to assume that Intel can just push prices down. We know that AMD's 690G chipset is used on 35 motherboards versus about 42 for Intel's 965. That suggests to me that someone plans to make 690 systems in the second half. Likewise, it looks like DTX and mini-DTX are successful and this too should keep Intel from being able to push down as much as it would like.

You can save a heck of lot on system cost with mini-DTX: smaller case, smaller power supply, smaller motherboard, less cooling, reduced shipping costs. This seems like a solid strategy. I figure AMD will have this in Q4 and then Intel will probably catch up in Q1. This should help reduce pricing pressure on the bottom.

On larger systems you can use 690G with DTX and this is still a better value over Intel based 965 systems.

I'm thinking the chipset and graphics division will be able to break even in Q3 and make a profit in Q4. I don't expect AMD in general to break even until Q4. Obviously, we get a better indication if AMD loses less money in Q3. My rough calculation would be that AMD would need to lose $310 Million in Q3 to be on track.

Christian Jean said...

Giant said...
Intel should commence a hostile takeover of AMD. Buy 50% of the shares by offering the shaderholders a 50% premium over AMD's current share price.

Once that's done they can terminate AMD's x86 license as per the terms of the agreement. Then AMD is finished.


There is more than meets the eye here! For starters, you've got to remember that AMD has been one of the largest patent producers every year for the last decade. I think they average like 3rd or 4th overall. Obviously not all patents were filed for x86 or processor architecture, but it just goes to say that AMD has a hell of a lot of technologies that Intel uses too. Don't forget that! If Intel were to terminate the cross-license, Intel would loose big time too... like x86-64 for example. Intel can use it, per their agreement, but AMD owns it.

Besides, its not really the 'know-how' or the 'implementation' that AMD is dependent of Intel, its the x86 instruction set. That is what is really vital to AMD!

Giant said...
You also have the advantage that both dual and quad core CPUs can be produced using the same dies. For AMD to do that would be quiet expensive, unless they have lots of Barcelona CPUs with two defective CPU cores.


I've never worked in a FAB before, so this may sound like a dumb question, but here goes...

Could a quad-core, which is found to be defective be cut in two in order to keep a dual-core? Or would this be too 'challenging'?

Fujiyama said...

Scientia,
I think that you are too optimistic about AMD future. There are no indicators that K10 is going to manage upcoming 45nm constructions and Nehalem next year.
And AMDers knows that so they are trying to create a few different and unique products hoping that it will draw attention.

I think that they are trying to create three different CPU lines:
- low cost, low ASP based on Athlon X2 and Fusion idea.

- Phenom/Barcelona large die and high ASP models

AMD understands more and more that IT CANNOT EARN MONEY ON CPU.
Now they try to sell the whole platform to OEMs hoping for higher ASP overall.
The main added value is good graphics inside chipset or CPU.
So selling 30$ worth GPU they want to upgrade ASP up to 160-200$.

This is main AMD problem UNSOLVED.
How to increase prices giving the same value to us. And faster-than-light Phenom cannot help with this. AMD was not able to take over more than 30% of the market having ultra-fast Opteron line. So 5-20% faster Barcelona
is not going to solve that problem.

This is main reason buying ATI - the brand and hope to get more than 30% with good platform coverage.

Scientia from AMDZone said...

Fujiyama

"There are no indicators that K10 is going to manage upcoming 45nm constructions and Nehalem next year."

I don't know what you mean. Shanghai will be 45nm in Q3.

"AMD understands more and more that IT CANNOT EARN MONEY ON CPU.
Now they try to sell the whole platform to OEMs hoping for higher ASP overall."


You're a bit late with this. AMD's first chipset was 750 for K7. AMD's biggest problem lately has been the lack of a good mobile chipset. This comes in Q2 2008. However, with the AMD factory stamp on ATI products, the 690G chipset has been a great success.

You are also forgetting AMD's efforts to trim system cost with DTX.

"AMD was not able to take over more than 30% of the market"

The number is actually 25% and it is because AMD only had one 200mm FAB. It still plans to hit 30% in 2008.

Scientia from AMDZone said...

axel

Look here for Intel's latest information.

Page 13 shows Intel's current production estimates for 45nm:

Q4 07: about 3% 45nm
Q1 08: about 12% 45nm
Q2 08: about 30% 45nm
Q3 08: crossover
Q4 08: about 70% 45nm

So, you can see that Intel will not have a lot of 45nm chips even in Q1. 45nm will start having real effect in Q2 at 30%. However, it doesn't get much higher since AMD should introduce 45nm in Q3. 45nm should have a real effect on servers in Q1 but 3% isn't going to make much of dent on even servers in Q4, maybe just the upper speed grades.

On page 15 you can see where Intel gives its lineup out to 2010 with Sandybridge. You seemed to complain that AMD was looking to far ahead with 2009.

Just for fun you might look at page 17 and see how many incorrect things Intel has listed. The lines that do not contain errors are: Technology, Total L2/L3 Cache, and Issue Width. However, Issue width is misleading because it implies no change from K8 to K10 just like no change from Core to Penryn. This is misleading because although K10's issue width did not increase it can issue SSE instructions twice as fast as K8.

Ho Ho said...

scientia
"Again, the supershuffle helps Intel a lot; AMD doesn't need it. "

How fast are shuffling operations on K8/10 compared to Core2/Penryn?


" Obviously, ICC would not be proper code (K8 currently takes a 20% hit)."

I remember seeing a link on this very same blog where I saw that ICC optimizes code for K8 a lot better than MSVC. That means performance difference between C2D and K8 was bigger with MSVC and smaller with ICC. I think it was some kind of encoder that was compiled on both compilers and ran on both CPUs. Unfortunately I can't remember where that link was.

Do you have any updates on how does ICC compile code for AMD? ICC10 would be especially interesting.


"You can save a heck of lot on system cost with mini-DTX: smaller case, smaller power supply, smaller motherboard, less cooling, reduced shipping costs. This seems like a solid strategy."

So everything is small and cheap? Including the CPU? If so then it doesn't seem to help with rising ASP too much.


Also how on Earth could AMD reach $100 ASP in Q4 with their dirt-cheap K8 CPUs when K10 will have neglible marketshare this year according to their own words? There is no way that graphics or chipset division could help a lot with profitability, after all even their graphics part earned quite a bit of losses last quarter.

Axel said...

Scientia

There is no reason for Penryn to gain 31% in SSE; it's SSE pipeline is nearly identical. This might happen with SSE4 in some cases but not in general cases.

Then would you mind conjecturing on the reason for the apparently huge gains in HL2 and fairly healthy gains in the other games?

Again, the supershuffle helps Intel a lot; AMD doesn't need it.

Can you explain or link?

abinstein said...

scientia -
"Page 13 shows Intel's current production estimates for 45nm:

Q4 07: about 3% 45nm
Q1 08: about 12% 45nm
Q2 08: about 30% 45nm
Q3 08: crossover
Q4 08: about 70% 45nm

So, you can see that Intel will not have a lot of 45nm chips even in Q1."


Disregarding what axel has said, I think AMD will have the upper hand if it can start 45nm production by 3Q08, when Intel has 45nm cross-over. If so, by the end of 2008 Intel will ship 60% 45nm while AMD close to 50% (according to its 65nm ramp speed).

However there are two "catches". First, AMD might not have 45nm in production until 4Q08. Second, AMD might have to slow down 45nm ramping if its finance doesn't improve.

My point is, it really depends, and nothing is set to certainty. If Barcelona and Phenom can improve AMD's ASP and revenue, and if its 45nm stays on-plan, then it's very likely for AMD to get the upper hand of Intel again by the end of next year. OTOH, if somehow AMD's financial doesn't improve or it slips on the 45nm plan, then it will keep lagging behind Intel on both technology and profit for a long time to come.

abinstein said...

"Could a quad-core, which is found to be defective be cut in two in order to keep a dual-core? Or would this be too 'challenging'?"

You don't need to "cut" the die, just leave the defective cores there, disable it/them, and sell the die as dual-core (or maybe tri-core?).

It's already been done on dual-core Athlon64, where dies with one defective core is sold as single-core Athlon64. For all practical matters it is the same, both power or performance wise.

abinstein said...

"Yeung's estimate of a loss of a $2.59 loss was only slightly worse than the consenus estimate of a loss of $2.48."

That would make him below the average analyst quality, wouldn't it? :-)


"And, he was just one month more pessimistic than the September actual availability."

The problem is that he seems to be manipulating the market with exaggerated estimates. We already knew Barcelona will be delayed, from early summer (Jun/Jul) to late summer (Aug/Sep). He came in to add one more month to the delay (Sep/Oct).

If this inaccuracy is not FUD, what is?

anonymous said...

scientia wrote:
Q4 07: about 3% 45nm
Q1 08: about 12% 45nm
Q2 08: about 30% 45nm
Q3 08: crossover
Q4 08: about 70% 45nm


Assume a 25% cost reduction due to die size reduction for the 65-->45 transition (accounts for additional cache, process costs). Then per your numbers, the net COGs reduction is:

Q4 07: 3/4%
Q1 08: 3%
Q2 08: 7.5%
Q3 08: 12.5%
Q4 08: 17.5%

Given that Q2 COGS rang in at ~4.6B, a 3% reduction is about $140M- hardly insignificant.

Also, remember that while the impact of 45nm COGS reduction takes a little longer to show up in improving GM, the fact that 45nm capital purchasing has tapered substantially will also result in a more immediate GM/NM boost.

AndyW35 said...

Scientia from AMDZone said

"AMD changed the L2 controller so that they could add more L2 cache. Some have suggested that this was a test bed for K10 since there is no indication that AMD is actually increasing the cache size on Brisbane."

And how much L2 cache does K10 have? Less than K8 at 90nm is the answer.

So that theory to me seems rather bogus. More likely it increased yields on 65nm Brisbane, but they were not prepared to admit it at the time.

The increased cache size was just a red herring it now seeme with hinsdsight.

Aguia said...

AndyW35,
K10 L2 also has 12 cycles.

New Cache Structure

The K8 65nm and K10 all have smaller L2 cache and lower speed comparatively to K8 130nm and 90nm. I wonder what happened too. Maybe the 65nm process was improved for lower power consuming and lower temps at the exchange of clock speed?

There have been reports the K8 will also be updated to accept HT3. So maybe the K8 design will keep being upgraded with some parts of K10 and the L2 cache plays a roll in that. Who knows?

After all Griffin is still a K8.

abinstein said...

"So that theory to me seems rather bogus. More likely it increased yields on 65nm Brisbane, but they were not prepared to admit it at the time."

The "larger" L2 that AMD referred to is actually the shared L3. Before L2 can return with a hit, K10 will need to check whether a modified cache line exists in L3, or whether to invalidate the cache line from L3. This will easily take two cycles.

I believe AMD extends the L1 miss penalty, together with some tweaks, first in Brisbane, which shows there's negligible performance penalty associated with it, anyway.

abinstein said...

aguia -
"New Cache Structure"

Nop, that picture is not accurate. The L1 cache in K10 has direct data path from the memory controller. Both L1 and L2 cache in K10 have direct control paths to and from the memory controller as well. The L3 does not get in-between, which by definition is how the victim cache works.

Ho Ho said...

abinstein
"There have been reports the K8 will also be updated to accept HT3. So maybe the K8 design will keep being upgraded with some parts of K10 and the L2 cache plays a roll in that. Who knows?"

Nobody of us probably knows for sure but playing around with K8 cache latencies and thinking it is just for testing how things might work for K10 is quite definitely wrong. For testing you have special testing chips, you don't run tests on production CPUs you intend to use for generating money.


Btw, does anyone have any idea how could AMD rise its CPU ASP to around $100 by Q4? If $100 is too much then how high could they realistically get? IIRC AMD didn't say they will get even in Q4, they said they would like to do so.

abinstein said...

Ho Ho -

First, you didn't quote me correctly.

Second, -

"For testing you have special testing chips, you don't run tests on production CPUs you intend to use for generating money."

Not even if the resulting design has better power and yield with negligible performance penalty?

Scientia from AMDZone said...

ho ho & axel

K10 got a 2X speedup too.

Current AMD Family 10h processors support two SSE logical/shuffle units, one in the FMUL pipe and another in the FADD pipe, while previous AMD64 processors have only one SSE logical/shuffle unit in the FMUL pipe. As a result, the SSE/SSE2 shuffle instructions can be processed at twice the previous bandwidth on AMD Family 10h processors.

Furthermore, the PSHUFD and SHUFPx shuffle instructions are now DirectPath instructions instead of VectorPath instructions on AMD Family 10h processors and take advantage of the 128-bit floating point execution units. Hence, these instructions get a further 2X boost in bandwidth, resulting in an overall improvement of 4X in bandwidth compared to the previous generation of AMD processors.

Aguia said...

Nobody of us probably knows for sure but playing around with K8 cache latencies and thinking it is just for testing how things might work for K10 is quite definitely wrong. For testing you have special testing chips, you don't run tests on production CPUs you intend to use for generating money.

Like Prescott or Willamette for example.

Not even if the resulting design has better power and yield with negligible performance penalty?

Axel said...

Scientia

That's impossible. AMD is on a 300mm FAB now; you can't put the genie back in the bottle.

Of course you can. Even if a company were to produce billions of CPUs a quarter, they can still have 0% market share if no one wants them. It looks like there'll be a glut of CPUs on the market in mid-2008 (if not already) with both AMD fully ramped on Fab 36 and Intel well into the 45-nm ramp. Both companies may have to write off millions in inventory, but overall market share will be determined by market preference.

You still haven't shown a logical argument for why AMD can price a 2.5 GHz Kuma the same as a 2.66 GHz Wolfdale. What benchmarks are out there to support that position? Remember, this is the desktop space we're talking about where K10's advantages in memory bandwidth and power efficiency aren't so important. SPEC_fp doesn't count. The early Cinebench and POVray previews are not promising for K10 despite 128-bit SSE processing.

I maintain that Wolfdale's SSE capability and power efficiency as evidenced by the HKEPC preview indicate that *for the desktop, workstation, and mobile markets*, Wolfdale is likely to bring more value to the table than Kuma will. Therefore I maintain that a 3.00 GHz Kuma is likely to be priced similarly to a 2.66 GHz or 2.8 GHz Wolfdale. Can you show otherwise instead of simply saying "that's wrong"?

In the server space K10 may very well bring more performance per clock than Penryn and perhaps even per watt, but I just don't see it for the other markets. If AMD cannot show that K10 has more value than Penryn, Intel can very well manage to put the genie back in the bottle unless AMD prices K10 very low. But then they hemorrhage into bankruptcy.

No, K10 will not be AMD's savior. Perhaps Bulldozer will but can they hang in that long?

abinstein said...

axel -
"I maintain that Wolfdale's SSE capability and power efficiency..."

Keep maintaining. As I said, you are just FUDing without any backup and will simply be proven wrong, like you will be on K10 SPEC score and countless other issues.


"Wolfdale is likely to bring more value to the table than Kuma will."

Have you seen Kuma at all? If quad-core Phenom 3GHz with 2MB L3 has only 120W TDP, Kuma 3GHz, without the L3, likely has just 50W TDP or less. This is including the memory controller. The HKEPC shows Wolfdale taking 59W at merely 2.33GHz, without the memory controller.


"Therefore I maintain that a 3.00 GHz Kuma is likely to be priced similarly to a 2.66 GHz or 2.8 GHz Wolfdale."

s/maintain/FUD


"Can you show otherwise instead of simply saying "that's wrong"?"

I just shown otherwise above. OTOH, can you show evidence instead of saying "I maintain"?


"No, K10 will not be AMD's savior. Perhaps Bulldozer will but can they hang in that long?"

Yes, I believe to you Intel's 64-bit Itanium, or its 5GHz processors are the true saviors. Definitely not AMD's Athlon, Athlon64, or (now) K10.

abinstein said...

Oops... I miss-typed Kuma's TDP estimate. It should've been 60W or less at 3GHz. Still better than Wolfdale's 59W at just 2.33GHz.

Purely for its better power-efficiency, I believe a Kuma at 2GHz will probably be priced comparable to Wolfdale 2.33GHz, and Kuma at 3GHz will likely have no competition from any Intel's processors until Nehalem.

Axel said...

abinstein

The HKEPC shows Wolfdale taking 59W at merely 2.33GHz, without the memory controller.

Clearly those HKEPC power figures are accounting for more than the CPU alone, as the E6550 is consuming 83W and its TDP is 65W. I would guess that it's CPU + NB or something like that. I would guess that the Wolfdale CPU alone in that preview is only consuming 45W or less at load.

Axel said...

Scientia

Yeung's estimate of a loss of a $2.59 loss was only slightly worse than the consenus estimate of a loss of $2.48. Of course, both of these were wrong since the actual loss was $1.09.

No. Their estimates were for total losses in 2007, not for Q2 alone. We have yet to see how this will play out. So far they seem to be on track.

There's so much misinformation here it's beyond comprehension. Baseless assertions that K10 will somehow save AMD's hide when all we have is Powerpoint slides. Never mind the fact that we have ACTUAL BENCHMARKS of K10 in action with Cinebench and POVray that don't impress. No, let's just dismiss actual benchmarks and believe AMD's marketing baloney.

enumae said...

Abinstein

I know that these are not desktop parts, but Intel (according to DailyTech) has an 3.33GHz Dual-Core at 65W.

Also looking at the article, and doing what you have done (split the power for a 3GHz Quad-Core), we could see a 40W 3.0GHz Dual-Core.

Axel said...

Abinstein

Remember Glen Yeung who predicted AMD will lose $2.59 per share...

Just to flesh out my previous comment, let's run some numbers. Yeung predicts a loss of $2.59 per share for 2007. AMD currently has about 550 million shares outstanding ($7.24 B market cap / $13.15 current price). For AMD to lose $2.59 per share in 2007, they have to lose about $1.4 billion for the year.

So far AMD have already lost $1.2 billion. It looks they will far exceed Yeung's estimate of losses for the year. The reason Yeung is quiet is because he didn't think AMD would do SO MUCH WORSE than what he predicted.

Scientia from AMDZone said...

axel

"No. Their estimates were for total losses in 2007, not for Q2 alone. So far they seem to be on track."

Oh, for the year's total I would agree.

"Baseless assertions that K10 will somehow save AMD's hide"

Baseless assertions that K10 is worse than what Intel will offer. Besides you really don't understand if you think it is just about K10. It is also about ATI's offerings and DTX and AMD's mobile offerings and about cost reduction as AMD ramps down FAB 30. It's also about faster clock speeds with each stepping.

"when all we have is Powerpoint slides."

It amazes me that you can claim that there is not enough information to say that K10 is good yet you also claim that the same lack of information is enough to prove that K10 is bad. Curious logic.

For the rest, you just keep repeating yourself. Basically, you have no information. However, you want to rely on the cinnebench and pov-ray scores because they would be bad.

You seem to forget that Intel was similarly guarded when it first showed C2D (strange how people forget). Anandtech was not allowed to do with the first C2D demo what they did with K10. No one was allowed to install benchmarks; they were only allowed to run what Intel had installed. Naturally, Intel only had installed the benchmarks that were favorable. Intel only allowed additional benchmarks when they thought they would look good. Anandtech had the same double standard with their pirated ES chips. Intel's ES chips would not have run on just any motherboard either but Anandtech implied that this was a special problem for AMD. Nonsense.

It is clear that the B0 stepping does not allow very high clocking. Intel had the same problem with their early Conroe's. It took two more steppings before Conroe could reach 3.0Ghz and another stepping after that for Kentsfield to reach the same speed.

We've seen that AMD has increased clock speed with B1 and B2 so clock speeds shouldn't be a problem in Q4 and Q1. With the improvements AMD has made to the architecture it is impossible that K10 is not substantially faster than K8. SSE instructions decode in half the time and shuffle is twice as fast. I assume you are now going to claim that AMD is putting false information in its Software Optimization Guide. That would be peculiar since that is the same guide that software developers are using today to make K10 software.

Yes, you want benchmarks but you also don't seem to mind if only the ICC compiler is used which gives K8 a 20% hit in performance. You would like to pretend that C2D has a 20% (or even 25%) lead when in reality the lead is probably closer to 12%. Stacking the deck in Intel's favor does not prove that C2D is better.

So, do you admit that if you don't have enough information to say that K10 is good you also cannot claim it is bad?

Do you also admit that using the ICC compiler is cheating in Intel's favor?

Do you admit that K10 is getting faster with each stepping just like C2D did?

And, do you admit that the Cinnebench test was something that Intel did not allow when they first demoed Conroe?

anonymous said...

scientia opineds:
Do you also admit that using the ICC compiler is cheating in Intel's favor?

At the risk of repeating myself, this is a red herring. The correct compiler for benchmarking is:

a) the one that commonly used apps are compiled with (probably MSVC, then either GCC or ICC)

b) the one you use to compile your own applications (only valid for hobbyists or software houses)

c) the one that gives the best results for your CPU (if you are submitting SPEC scores)

PGI might be the greatest thing since sliced bread, but if their market share is low, they aren't really relevant to a discussion about benchmarks- unless you are in a very small minority of users.

My 2 cents.

Axel said...

Scientia

So, do you admit that if you don't have enough information to say that K10 is good you also cannot claim it is bad?

Until we see post-release benchmarks no one can claim anything with certainty. What we do have is a couple covertly obtained pre-release benchmarks of both K10 and Penryn. This is certainly more solid and objective than any material produced by AMD, such as Powerpoint slides or the K10 software optimization guide. Obviously we cannot draw firm conclusions yet, but the indications are there.

The current K10 software optimization guide clearly describes on pages 7-8 that to get the full benefit of the 128-bit SSE units, the code should be optimized in specific ways. Some of these optimizations may not apply to Conroe / Penryn. Now can you take a wild guess about which processor developers are more likely to optimize code for: K10 or Conroe/Penryn? This may explain why K10 didn't show much improvement over K8 per clock for Cinebench & POVray. I certainly haven't yet seen a logical explanation from you as to why this occured.

Nor have I yet seen a logical explanation from you as to why Half-Life 2 performance improved by 31% per-clock from Conroe to Penryn, without SSE4 code. How about taking a stab at that one?

So if you can't explain these data points away in a reasonably logical fashion I guess you can just ignore them and pretend that the K10 Software Optimization Guide is the holy grail of evidence for K10 performance improvements. You can ignore the fact that Cinebench was run by AMD themselves. Well you'll have to forgive me if I take a more skeptical stance towards claims about K10 SSE performance. I see K10 perhaps catching up to Conroe, but not to Penryn.

Now can you offer any logical counterpoint to this besides:
- "You're clearly wrong".
- "But the K10 Software Optimization Guide clearly shows...."

bk said...

Dr. Yield
"PGI might be the greatest thing since sliced bread, but if their market share is low, they aren't really relevant to a discussion about benchmarks- unless you are in a very small minority of users."

By this logic all benchmarks should be using Visual Basic.

If PGI does an equally good job at optimizing code for both Intel and AMD, then it is a much better indicator of performance than a compiler that is slanted one direction.

Please someone try to argue that Intel's compiler is not biased towards Intel chips.

Scientia from AMDZone said...

Well, I'm trying to find something that might tell what share PGI has on the desktop. I know that in HPC they are #1. This is backed up by other statements such as this one from Hiperism Consulting:

It has been our practice to examine all available compilers for Linux systems, but we have tended to confine attention to the commercially supported ones. Even amongst these the main interest is in the compilers in use by the AQM community and these include those from Absoft, the Portland Group, and Intel.

However, this statement from Hiperism is also interesting:

Precise performance measured comparisons show that even minor compiler version upgrades of a specific compiler (say from x.0 to x.1) can show performance boosts in the range 12%-22%

This is for parallel code which is not a factor for single core but could be a big factor on quad core.

Scientia from AMDZone said...

This is from Digit Life 2005:

The Portland Group product is rather popular, due to OpenMP and MPI support in particular. That's one of the first commercial compilers with AMD64 support.

lex said...

BK getting closer and closer
http://www.marketwatch.com/news/story/amd-taps-debt-market-again/story.aspx?guid=%7B0EE99395%2D7D4F%2D452C%2DA5A5%2DCDA9ED74689D%7D&dist=TQP_Mod_mktwN

5.75% are they joking. AMD is a company that needs to be paying junk bond returns. They aren't going to be in business in 2012.

They don't have enough market share to generate enough cash to develop and ramp 45nm and 32nm between now and 2012. What a joke, is it no wonder the credit markets were so bad in forecasting risk and are in the pickle they are in now. Them jokers at Morgan Stanley will be pulling a Goldman in 6 months when AMD goes belly up.

You read it here first. Phenom and Barcelona prices are going to do a Opeteron like dive in Q1'08 as Penrym ramps and crushes them. Then they will be in celeron space when Nehalem appears later in 2008.

BK in 2008..

Ho Ho said...

Still no comments how could AMD significantly increase ASP by the end of the year? Also nothing about ICC vs other compilers besides some old stories. Scientia, have you found that link I was talking about that showed that ICC generates great code for K8? Can you link to something that compares somewhat newer compilers, like 9.1 or 10?

Scientia from AMDZone said...

lex

"They don't have enough market share to generate enough cash to develop and ramp 45nm"

Wow, you couldn't be more wrong. 45nm test wafers are running now so there isn't much development left to do. Also a good bit of the 45nm tooling is being bought in 2007. Also, the 45nm ramp is easier than the 65nm ramp for two reasons: First, AMD has the 45nm tooling in a timely fashion instead of having to wait for a new FAB as it did with FAB 36. Secondly, FAB 36 startup required cross iteration from both FAB 30 and East Fishkill. This is considerably easier with 45nm as the iteration is only East Fishkill to existing 65nm.

" and 32nm between now and 2012."

Your timeline is off. 32nm is 2010. 2012 would be 22nm. And, 22nm is up in the air even for Intel.

"You read it here first. Phenom and Barcelona prices are going to do a Opeteron like dive in Q1'08 as Penrym ramps and crushes them. Then they will be in celeron space when Nehalem appears later in 2008."

Yes, I'll put this one in my list of Absurd Predictions along with roborat's claim that AMD will be completely outsourced in 2008. There will be plenty of crow to go around.

Intel's volume ramp is one quarter behind AMD's:

Q3 - AMD K10 intro
Q4 - AMD server, Intel Penryn intro
Q1 - AMD desktop, Intel server
Q2 - AMD mobile, Intel desktop

Good desktop volume for Penryn isn't until Q2 which is when AMD will be ramping its new mobile cpu and chipset. Centrino will get its first real competitor. In Q3 AMD should be beginning to produce Shanghai. In Q4 Intel should have Nehalem but that is a double edged sword. While Intel gains in IMC and CSI Intel loses in monolithic die size. Intel also loses because the cpu will have to power both CSI and IMC.

You obviously missed the part where they said that AMD should now have enough cash to get through 2007 and will only need more cash in 2008 if they keep losing $500 Million every quarter.

Fujiyama said...

AMD should now have enough cash to get through 2007 and will only need more cash in 2008 if they keep losing $500 Million every quarter.

Scientia - look at AMD chart and answer the question - why investors don't believe that in a few weeks AMD is going to "change the game" as Randy Allen says?

Hector is dreaming about the end of 2005 where X2 was a hero. But it is not going to come back.

1. Did you calculated the number of quad-cores AMD can produce with 65nm? How they are going to match 150$/45nm worth Intel quads?

2. K10 is going to reduce looses, increase ASP and give small profit but what then?
How AMD is going to get back 1169 mln they lost in Q1 and Q2???
If everything goes right they earn 130mln per quarter! So we are going to wait 2 YEARS UNTIL THE GAP IS GOING TO BE FILLED.

This is not a business, this is some kind of survival game.
And Hector with the rest of the guys should go.

Aguia said...

Well bad news...

Advanced Micro's Ruiz Says Chip Delay Limits Gains

How difficult was to make just one Quad K8 one year ago.
And why not start to manufacture, single/dual core K10 first.

K8 quad was enough to crush Xeon and still give a good fight to Core 2 Quad.

Scientia from AMDZone said...

Ho Ho

"Still no comments how could AMD significantly increase ASP by the end of the year?"

Just from shifts in server chips, AMD's ASP should come up $10. I don't know if there are other cost reductions that AMD has in mind that would reduce the required ASP needed to break even in Q4.

" Also nothing about ICC vs other compilers besides some old stories."

I've only seen one direct comparison between ICC and PGI. That is not an easy webpage to find again.

" Scientia, have you found that link I was talking about that showed that ICC generates great code for K8?"

Oh, I'm sure it does compared to Microsoft but I doubt that is the case compared to PGI.

" Can you link to something that compares somewhat newer compilers, like 9.1 or 10? "

Probably not. Review sites like Anandtech, Toms Hardare, Tech Reports, and Xbitlabs haven't seemed very interested in comparing PGI. Odd considering how many sites went out of their way to test Hyperthreading on P4.

Ho Ho said...

scientia
"Just from shifts in server chips, AMD's ASP should come up $10."

Will AMD finally switch from 90nm to 65nm for Opterons? From what I see all their Opterons are still at 90nm. If they intend to ramp K10 significantly any time soon I doubt it would make too much sense to try to produce 65nm Opterons for a few months. Though it wouldn't be the first weird thing they have done during the last year.


"I've only seen one direct comparison between ICC and PGI. That is not an easy webpage to find again."

Can you remember the ICC version they used there?


"Oh, I'm sure it does compared to Microsoft but I doubt that is the case compared to PGI."

Well, I may doubt the opposite. I wonder who will need to bring links to prove itself :)


"Odd considering how many sites went out of their way to test Hyperthreading on P4."

Could be because HT was an interesting thing and customers might have benefited from it. What do you think how many home-users even know what "compiler" means?

Aguia said...

What do you think how many home-users even know what "compiler" means?

And how many do you think know what Hyperthreading is?
If it’s even enable or disabled in their computer?

How many do you think know what AA or AF is in their GPUs?
How many enable these technologies on their GPU and how to enable them?
How many people benefit from these technologies?

Ho Ho said...

aguia
"And how many do you think know what Hyperthreading is?"

Advertising tells (told) them it increases performance, this is all customers care about. Also they will get a bit better responsiveness compared to singlecore CPU.


"If it’s even enable or disabled in their computer?"

I've yet to see a PC with HT disabled by default.


"How many do you think know what AA or AF is in their GPUs?"

Again, advertising tells them it increases image quality and they can see it themselves.


"How many enable these technologies on their GPU and how to enable them?"

Usually there is a setting in games called something like "maximum quality" or at least there is a descripton besides AA slider that it'll increase quality.


"How many people benefit from these technologies?"

I've just listed the things. I could ask a similar list why should regular people care about different compilers. Vast majority will not run it even once in their life.

Scientia from AMDZone said...

Fujiyama

"Scientia - look at AMD chart and answer the question"

What chart?

" - why investors don't believe that in a few weeks AMD is going to "change the game" as Randy Allen says?"

I have no idea what you are talking about. What investors, where?

"1. Did you calculated the number of quad-cores AMD can produce with 65nm?"

Roughly, 31 Million.

" How they are going to match 150$/45nm worth Intel quads?"

Are you trying to claim that Intel's average quad core ASP will be $150 in 2008?

"How AMD is going to get back 1169 mln they lost in Q1 and Q2???
If everything goes right they earn 130mln per quarter!"


Where did you get that number from? AMD should be able to earn $500 Million a quarter.

Unknown said...

"How many do you think know what AA or AF is in their GPUs?"

It takes a high end GPU to run a game at decent resolutions (1280x1024 and up). By 'high end' I'll refer to the X1950 XT/XTX and HD 2900 XT for ATI/AMD and the 8800 series for Nvidia.

The people who ATI/NVIDIA market these high end cards (with a minimum price of $200 USD usually) to certainly do know what AA/AF do.

We're not talking about Bob buying an $800 Dell computer that came with a GeForce 7200 GS or something similar. A card like that was never designed to run modern games with AA/AF and it's doubtful the buyer of such a card would have a clue as to what AA/AF are.

Scientia from AMDZone said...

Aguia

"Well bad news..."

Yes, the same bad news; nothing new.

Advanced Micro's Ruiz Says Chip Delay Limits Gains"

They've been saying that for awhile.

"How difficult was to make just one Quad K8 one year ago."

I would have to guess that they decided to put their efforts into K10.

"And why not start to manufacture, single/dual core K10 first."

You have it backwards. Kentsfield is just two Conroe's stuck together but K10 is a monolithic die. You don't get dual core first. The die has to be remapped to have a true dual core K10.

"K8 quad was enough to crush Xeon and still give a good fight to Core 2 Quad."

That is true if AMD had enough lead time and resources to do it. It is also very possible that AMD profiled this configuration and decided against it. Two X2's in one package is much less than optimal because of HT 1.0 and full width IMC. You can handle this much better with HT 3.0 and the more sophisticated controller in K10. Most likely AMD decided not to do it and then K10 was delayed. Hindsight is always clearer.

Aguia said...

We're not talking about Bob buying an $800 Dell computer that came with a GeForce 7200 GS or something similar. A card like that was never designed to run modern games with AA/AF and it's doubtful the buyer of such a card would have a clue as to what AA/AF are.

Like the ones that I know that have 6600GT and 7600GT cards for Internet and Word...
Most people don’t know anything.
The ones that came to me to advise them to buy X1950GT/PRO for their "old" P4 with AGP just want one of those because they know it’s faster, not because of AA/AF or even power consuming.

Scientia from AMDZone said...

Ho Ho

"Will AMD finally switch from 90nm to 65nm for Opterons?"

There's nothing magic about 65nm. As FAB 30 draws down and FAB 36 comes up, 65nm will replace 90nm. Of course, with the current low demand it wouldn't suprise me if AMD phased out 90nm more quickly.

"If they intend to ramp K10 significantly any time soon"

They are ramping now.

"Can you remember the ICC version they used there?"

No. I think it was last year.

"Well, I may doubt the opposite. I wonder who will need to bring links to prove itself :)"

Portland Group is very serious about its compiler. Intel's efforts are split between Itanium and x86. I would be quite surprised if Intel managed to get more than 2% ahead of PGI.

"Could be because HT was an interesting thing and customers might have benefited from it. What do you think how many home-users even know what "compiler" means? "

You have a point. And, yet. I've seen websites compare VCC, ICC, and GCC.

Aguia said...

Ho ho

If advertising solves the problem, then did you know that when I worked in a computer store in 1995/96 most of the people the came to the store with their computer with their amazing Sound Blaster AWE32, almost all of the customers Sound cards had the default FM audio enabled instead of the better sound quality General Midi enabled.

The advertising did make them buy the product but didn’t enable the Wave Table.
But it must be like you say advertising “automatically” enable HT and other features in all computer bios/systems.

Is DX9, is DX10, is AGP, is PCIe? Those are their requirements, even those "basic" requirements some people still fail (Like buying PCIe card to AGP mobo). At the best they read FPS charts but not features/specifications.

Ho Ho said...

scientia
"There's nothing magic about 65nm."

Yes but for some reason AMD still doesn't have any 65nm Opterons. Have they said when will they start selling those? For 90nm desktop parts you said they are on 90nm because they are the high-end CPUs and there are relatively few of those sold. Somehow I don't think the same excuse could be used for Opterons.


"They are ramping now."

Yes and my point was that bringing out cheaper, high-clocking and less power consuming K8's they will be fighting with their own dualcore K10's.


"No. I think it was last year."

ICC 9.1 came out early last year, IIRC. I know that the free Linux version was availiable a few months later in August. I know that 9.0 had a few "problems" with non-Intel CPUs. I haven't seen anything on 9.1 and 10.0 but it might be because I haven't looked hard enough.


"Portland Group is very serious about its compiler. Intel's efforts are split between Itanium and x86. I would be quite surprised if Intel managed to get more than 2% ahead of PGI."

Actually quite a lot of architecture specific optimization is simply done by modifying a few cost parameters. Most of the hardest part, optimizing the code (vectorizing, using SSE3 instead of regular ops, rearranging stuff for less dependencies etc), is done at much higher level. I'd say that quite a lot of code is being shared over Itanium and x86(64).

Also Intel has been working really hard on software side during the last few years, mostly to make it easier to write multithreaded code. Quite a lot of it depends on compiler and I'm quite sure their OpenMP support is one of the best in industry, if not the leading. With more and more cores in PC's that will soon be much more important than optimizing single threads.


"And, yet. I've seen websites compare VCC, ICC, and GCC."

So have I, though I can't remember more than two with MSVC and one was that one I talked about and the other was >4 years old. Can you link to some newer ones? How many of those pages were dedicated for Joe Average? Most I've seen are for web server owners and Linux users. Neither havs too much impact on market. From what I understood aguia was trying to make a point that average users don't know or understand things like HT or AA.

Btw, are there any half-decent comparisons made with PGI vs others, preferrably by someone else but Portland Group itself?


aguia
"The advertising did make them buy the product but didn’t enable the Wave Table."

Thanks for proving my point :)
Advertising is very important if you want to sell your stuff. Once it is sold customers often don't bother to make sure it is even working for them. E.g compare the number of sold SLI/CF motherboards to the nubmer of dual-GPU systems in use.


"But it must be like you say advertising “automatically” enable HT and other features in all computer bios/systems."

I've never said that it can enable all other features. When we specifically talk about HT then I can say that on the tens of PC's I've had contact with all had it enabled by default. You had to disable it manually to make it stop working.

Aguia said...

From what I understood aguia was trying to make a point that average users don't know or understand things like HT or AA.

You understood perfectly! :)

abinstein said...

Ho Ho -
"AMD still doesn't have any 65nm Opterons."

It doesn't make sense for AMD to make 65nm Opteron. By the time these new stepping chips pass QA the K10 will be out shortly.

Ho Ho said...

aguia
"You understood perfectly! :)"

They don't know what they mean but they do know that advertising told them it is something good. Also they can see with their own eyes that HT can improve speed and responsiveness and AA increases image quality.

For the record, the biggest change I saw with going from HT P4 back to non-HT one was that Eclipse suddenly became extremely laggy. Every time there was some CPU intensive thing going on entire GUI froze. With HT (and dualcore) GUI thread can continue running in parallel.



abinstein
"It doesn't make sense for AMD to make 65nm Opteron. By the time these new stepping chips pass QA the K10 will be out shortly."

Yes, that was exactly my point. Scienta was talking about "shifts in server chips". AMD itself has said K10 has almost no impact on this year financials so I logically assumed he was not considering selling a few K10's and next best thing that could lower costs was move to 65nm.

Scientia from AMDZone said...

Ho Ho

"For 90nm desktop parts you said they are on 90nm because they are the high-end CPUs and there are relatively few of those sold. Somehow I don't think the same excuse could be used for Opterons."

Are you joking? Server cpu's are less than 10%; FAB 30 is still more than 33% of capacity.

Ho Ho said...

Ok then. Still I can't understand what you meant by "shifting". As you said, server chips make up less than 10%. How high must their prices go to rise overall ASP by $10? I don't think AMD can somehow sell Opterons at $100 higher price point than it has during last few months.

Aguia said...

FAB 30 is still more than 33% of capacity.

That’s why the rumors or more than certain that AMD will release 89W 3.0Ghz CPUs and the 3.2Ghz CPUs (6400+). There is a possibility (also rumor) that they can hit 3.6 GHz if they want (still need). For 90nm that’s amazing.

Ho Ho said...

aguia
"For 90nm that’s amazing"

It really is amazing. Though I wonder why are they improving 90nm this much if it would be much better to put that money into improving their 65/45nm technologies. Especially considering 90nm won't be living long.


According to this page AMD has to pay back around $0.9B every year until 2011. This is my own calculation that doesn't take intress into account. With intress that can be much higher. Where on Earth can AMD get such money? I don't think lessening their ASP's with Fusion and DTX helps all that much.

Ho Ho said...

Forgot to mention that after 2011 they still have loans to pay back, just that one major one should get repaid by then and in 2012 two more. Longest lasting debt of $2.2B should get paid back around 2015.

Also analysts expect that it takes until 2009 until AMD gets back to positive. Even I expected it to happen sooner, I expected it to happen aroun Q3 2008. Though combining $1B loan payments per year with >$5B in capital expenditures it doesn't take a genius to figure out how difficult it is to make profits.

Scientia from AMDZone said...

Ho Ho

"AMD itself has said K10 has almost no impact on this year financials "

You've just used the "Sweeping Generalization" fallacy. It is true that K10 will not bring in much revenue in 2007. However, the server segment is AMD's smallest volume segment which is currently less than 5% of total volume. There will be enough K10 Opterons in Q4 to raise the server ASP.

"so I logically assumed he was not considering selling a few K10's and next best thing that could lower costs was move to 65nm."

Moving Opteron to 65nm would lower costs but would not raise the ASP. In Q4 you will see AMD's server ASP and volume both rise. Realistically, the server ASP could go up 10-20% however the server chip volume could as much as double. Doubling by the end of the year is quite unlikely but that still leaves a lot of room for a big increase.

Scientia from AMDZone said...

Ho Ho

"It really is amazing."

since Intel can't do it.

"Though I wonder why are they improving 90nm this much if it would be much better to put that money into improving their 65/45nm technologies."

They are not putting that moeny into 90nm; they are putting it into 65nm. As AMD improves the 65nm process there is also a nearly free bump for 90nm. This is AMD's normal way of operating.

"Especially considering 90nm won't be living long."

True. 90nm production should stop in Q2 08.

"I don't think lessening their ASP's with Fusion and DTX helps all that much."

Fusion and DTX do not reduce ASP. Both of these reduce system costs. This could even allow a higher ASP.

Unknown said...

since Intel can't do it.

Why would Intel need to improve it's 90nm process like AMD is doing? AMD needs it's 90nm fab. It's 65nm process maxes out at 2.6Ghz for Brisbane but 90nm goes to 3.2Ghz. All of Intel's x86 CPUs are at 65nm. 90nm is used for (mainly) chipsets and the Itanium.

Scientia from AMDZone said...

giant

Intel has no reason to develop 90nm any further. But you are missing the point. If AMD's 90nm process is capable of hitting 3.2Ghz then what kind of development potential do you think the 65nm process has? This should allow K10 speeds to keep increasing.

enumae said...

Scientia
This should allow K10 speeds to keep increasing.


The question should not be whether they will continue to improve, but how quickly it will improve?

And also if those improvements can be made before transitioning or shrinking to 45nm?

Ho Ho said...

scientia
"There will be enough K10 Opterons in Q4 to raise the server ASP."

I think I can agree with that. Though it will also mean that there is no way AMD could increase its ASP enough this year to break even as desktop K10 will not have almost any impact on it. Can you please tell us how could AMD break even in Q4?

"True. 90nm production should stop in Q2 08."

Will all server CPUs be K10 based by then?


"Fusion and DTX do not reduce ASP."

Say we have $100 CPU and $50 GPU. When we put the two together in one die it'll cost around $125. When sold separately they would have made more money, wouldn't they? Of course those numbers are absolutely random but point remains.


"If AMD's 90nm process is capable of hitting 3.2Ghz then what kind of development potential do you think the 65nm process has?"

I somehow doubt AMD can improve its 65nm too much within 1.5 years or so until it gets replaced by 45nm. Of course there is a chance of 45nm being delayed, then it might have a point. Also if current use of 90nm is of any example then it is possible that mobile CPUs will use 65nm longer than others.

Btw, 90nm will be used for almost three years by AMD. 65nm seems to be used for half the time. Won't it increase costs thanks to producing relatively less CPUs with the same research+tools? Take NV for an example, they don't move around production nodes nearly as fast as ATI does and manage much better.

Ho Ho said...

If it is unclear then those prices I made up are retail ones, not manufacturing cost.

Scientia from AMDZone said...

Ho Ho

"Can you please tell us how could AMD break even in Q4?"

I already told you that I didn't know. I'm not sure of what AMD's scenario would be to break even in Q4.

"I somehow doubt AMD can improve its 65nm too much within 1.5 years or so until it gets replaced by 45nm."

They have been getting one bump a quarter.

"Btw, 90nm will be used for almost three years by AMD. 65nm seems to be used for half the time."

90nm would be gone now if it weren't for FAB 30. If you look at the introduction time from 90nm to 65nm then it is about 2 years. It is 18 months from the introduction of 65nm to 45nm. Then 18 months from 45nm to 32nm.

" Won't it increase costs thanks to producing relatively less CPUs with the same research+tools?"

Somewhat but AMD is trying to catch up.

"Take NV for an example, they don't move around production nodes nearly as fast as ATI does and manage much better."

We'll see if that lasts.

InTheKnow said...

If you look at the introduction time from 90nm to 65nm then it is about 2 years. It is 18 months from the introduction of 65nm to 45nm. Then 18 months from 45nm to 32nm.

I've said it before and I'll say it again. Crossover is a far more meaningful metric. It accounts for ramp, factory loadings, and process maturity. Do you happen to have the crossover dates for the nodes in question?

They have been getting one bump a quarter.

Yeah, but each bump is harder than the last. I wouldn't want to bet on the pace continuing.

Unknown said...

I suspect that it will. Nvidia bet an awful lot on TSMC's 130nm low-k process back with the NV30 GPU. When it became apparent that TSMC's low-k process wasn't ready they pressed on without it, using a regular 130nm process. They ended up with the awful mess that was the Geforce FX. They lost the performance crown, and a lot of sales, to Ati because of that mistake. Radeon 9700/9800 vs. the FX 5800 Ultra 'dustbuster' anyone?

I've still got my 9700 Pro video card, still running great. :)

Since then Nvidia has always been conservative with process technologies; that's why G80 is still produced at 90nm.

Ho Ho said...

scientia
"They have been getting one bump a quarter."

How so? To my knowledghe there are no new 65nm dualcroes released since first ones were released December last year.

This certainly isn't so for 90nm either. I believe FX models are where the fastest CPUs are released the earliest.
2.6: Jan 9, 2006
2.8: May 23 2006 (~1.5q)
3.0: Nov 30 2006 (~2Q)
3.2: >3q?

Did I miss something or did you think of some other CPUs?

Also with 2GHz introduction and 1 bump per quarter they could get from 2GHz to 3.2GHz quads in 1.5 years. Of course that number is randomly taken from wall and it doesn't show anything. If that 1 bump per quarter would be true then K10 wouldn't be all that fast by the end of its life. Of course there have been rumours about extremely good revision but I wouldn't hold my breath on it.

Anyone remembers similar rumours in Q1? Anyone remembers that those rumours appeared just before AMD took a loan? Guess what, it basically just took another loan shortly after those new rumours began. I hope this is simply a coincidence, I wouldn't want to loose my faith in AMD.


"We'll see if that lasts."

It has lasted since G40 when NV started to take marketshare from ATI and they have reported their best quarters yet. I somehow doubt it'll change any time soon.


giant
"They ended up with the awful mess that was the Geforce FX"

Performance wise FX wasn't really that great. Of course one can say that they kind of missed with targeting DX9.0b and ATI got it better. Though no wonder if that DX version was basically written down from R300 specs.

Aguia said...

So AMD is really going after 3.2Ghz (X2 6400+):
AMD Akiba Event showcases Phenom 3GHz processor and roadmap

The "new" processor is able to OC up to 3.6Ghz on air, but I can’t find that link no more sorry, at that time I thought it was FUD...

Ho Ho said...

Working link

3GHz with heatpipe cooler. Scientia calls those "watercooler-equivalent" and sais that shows pretty much nothing about what speeds could CPUs be released :)

Aguia said...

I was still talking of K8 ho ho.

But that type of cooling is not bat if it’s just for the FX and for the EE. After all those processors normally cost 999$...

Or don’t bring any cooler at all, since the people who buy those generally want to choose the cooler.

Scientia from AMDZone said...

Ho Ho

"How so? To my knowledghe there are no new 65nm dualcroes released since first ones were released December last year."

You are leaving out the 2.7Ghz 5200. I assume now that 90nm is moving up to 3.2Ghz that the 3.0Ghz part will be downrated to 89 watts and the 2.8Ghz part 65 watts. I would expect the 2.8Ghz part to be replaced by 65nm soon. However, you are right, that would be slower than one bump per quarter.

"Also with 2GHz introduction and 1 bump per quarter"

Now you are talking about something completely different. K10 will be 2.0Ghz at launch however it will not only increase at one bump per quarter.

There are several questions about this though and unfortunately the information is not available. It is possible that AMD's good 65nm chips are being sold as BE chips. It is likely that the top two speed grades are low enough volume that AMD doesn't need 65nm. It also appears that K10 has gone through more steppings than Brisbane.

The K10 chips that will be available this quarter B0 stepping were started last quarter. The sample chips that were demoed B2 stepping won't be available until later.

"Of course there have been rumours about extremely good revision but I wouldn't hold my breath on it."

AMD demoed B2; this is more than a rumor. The only question is whether the B2 chips will make it out sooner than Q1. However, even the B1 stepping is able to clock to about 2.5Ghz.

Scientia from AMDZone said...

Ho Ho

"3GHz with heatpipe cooler. Scientia calls those "watercooler-equivalent" and sais that shows pretty much nothing about what speeds could CPUs be released"

Another Ho Ho strawman argument. What I actually said was that a high grade air cooler like a Zalman or Coolermaster was equivalent to water cooling four years ago. The 3.0Ghz demo did not use any high grade cooling. The consensus of people who saw the demo was that it was a 125 watt chip.

Every indication is that the 3.0Ghz demo is indicative of processor performance in Q4 or more likely Q1.

Axel said...

Digitimes suggests initial Barcelona pricing. $390 for a 2.0 GHz Barcy, 95 W TDP.

Now Clovertown Xeon currently sells on Newegg for $370, for either the 1U or 2U version with a TDP of 80 W. In addition, today Intel introduced a low-voltage 2.0 GHz Clovertown with a 50 W TDP.

Apparently AMD figure that Barcelona will not significantly outperform Clovertown per-clock or per-watt in most server apps, else they would price it higher as they desperately need to raise their overall ASP. If K10 can't outperform Clovertown in the server space where it has big supposed advantages in memory bandwidth, floating point performance, and power efficiency, what hope does it have of competing against Conroe or Penryn in the desktop, workstation, and mobile spaces where those advantages are not as significant and SSE comes into the picture as well?

As I see it, might as well put a fork in AMD.

Axel said...

Now Clovertown Xeon currently sells on Newegg for $370

I need to amend this. I just looked up the tray price of the 2.0 GHz Clovertown and it's $316. So 2.0 GHz Barcy is priced some 23% higher than 2.0 GHz Clovertown. So this implies that Barcelona will outperform Clovertown to some extent per clock or per watt, but not by a large margin.

So then Barcy is probably roughly equal to Kentsfield in the desktop & workstation spaces per-clock and likely slower than Penryn in those same spaces.

It's all speculation until release, but the indications are starting to firm up.

Ho Ho said...

scientia
"You are leaving out the 2.7Ghz 5200"

When will/was it released? I couldn't see it on either Wikipedia or AMD cpu finder.


"K10 will be 2.0Ghz at launch however it will not only increase at one bump per quarter."

Only time will tell.


"AMD demoed B2; this is more than a rumor."

Yes, they did. For all that we could know it might have been >>120W and overheating even with that heatpipe thing. Of course it als could have been different.


Also didn't Intel demo some rather high-clocked 45nm CPUs in Q1 this year? Also I remember them running some game on a 65nm 3.4(6?) GHz quadcore and as we can see no such CPU will be released. My point is that demos are nice but only time will tell if those demoed things will actually be sold.


"However, even the B1 stepping is able to clock to about 2.5Ghz."

Simple question, how can you prove it? AMD has also said that it'll release 2.5GHz quads in Q2 but it didn't.


"The 3.0Ghz demo did not use any high grade cooling"

It was certainly much better than regular copper core-aluminium fin things Intel uses on its 130W+ Netbursts.


"The consensus of people who saw the demo was that it was a 125 watt chip."

Well, that's what people might think but as with pretty much anything that conserns K10, only time will tell if it really is so. I surely have my doubts in that. Otherwise AMD would release so low clocked CPUs.

If I were in the managment of AMD I'd sell those couple thousand 3GHz ones at around $2000 for the ones who might want them. Saying that they don't want to sell high clocked CPUs because they don't have a whole lot of those is kind of silly in my oppinion. If you have a few extremely good CPUs then sell them at extremely high prices, it's not like you'd loose or anything as you had sold that for much cheaper anyways.

Of course there goes a line between when it is useful to search those good CPUs but if AMD is to believe and those demoed 3GH ones didn't need a whole lot of searching then I see no problems with selling a few higher clocked ones at day one. Though there is always the possibility that it really wasn't all that simple to find those 3GHz ones ...


"Every indication is that the 3.0Ghz demo is indicative of processor performance in Q4 or more likely Q1."

My bet is on Q2/3, if AMD gets lucky.


axel
"So this implies that Barcelona will outperform Clovertown to some extent per clock or per watt, but not by a large margin."

Either that or AMD simply puts the price to the level when it would sell all the CPUs and earn maximum profits. Price is the last thing to use as performance metric, especially for products that aren't in great supply

Axel said...

Ho ho

Either that or AMD simply puts the price to the level when it would sell all the CPUs and earn maximum profits.

Remember that with the initially low volumes of Barcy, AMD will almost surely sell all they can make especially since Barcelonas are drop-in replacements for existing dual core Opterons. AMD will realize maximum revenue by pricing the first batch of Barcys as high as they can while still being price competitive with Intel based either on performance per clock or per watt. Hence, price should be a good indicator of relative performance.

Now this may change in mid-2008 when it appears there'll be a glut of CPUs on the market with Fab36 fully ramped and Intel churning out 45-nm from four fabs. At that point either company may try to undercut the other on price in order to reduce inventories. This is, of course, a no-win situation for AMD unless K10 wipes the floor with Penryn. The few indications we have so far of performance, along with the suggested initial Barcelona pricing, indicate that K10 will do no such thing and that AMD are in real trouble.

Mo said...

Sci,

Why don't you take your Myth debunctions and post em on your blog so we can comment on them?
obviously most who'd like to comment on them are Banned but a mod who's fingers are always itching to press the ban button..

Just saying you're get much more response here than at that place.

Unknown said...



Also didn't Intel demo some rather high-clocked 45nm CPUs in Q1 this year? Also I remember them running some game on a 65nm 3.4(6?) GHz quadcore and as we can see no such CPU will be released. My point is that demos are nice but only time will tell if those demoed things will actually be sold.


That was a 3.33Ghz Yorkfield CPU. The current rumors are that this very CPU will launch in Q4 as the "Extreme Edition" of the new Yorkfield and Wolfdale lineup for desktops. Then the other quads and dual cores for desktops should arrive in early Q1.

Scientia from AMDZone said...

ho ho

Phenom 3.0 demo surrounded by FUD


"The next thing is that it was either heavily cooled so it could be overvolted to the moon and clocked absurdly high. This again is not true. It was running an aluminium HSF with a few heat pipes and a bog stock fan.

You can see that the fins are not copper, and other photographs confirm this. What does this mean? The CPU was air cooled, and you can't pump all that much wattage through a simple copper/aluminum HSF."

So, apparently your eroneous assumption is that this heatsink is comparable to something like a Zalman. It's not even close. The heatpipe in the picture is more like what I have on my notebook. I can safely say that that setup is not dumping 150 watts.

Scientia from AMDZone said...

Mo

"Why don't you take your Myth debunctions and post em on your blog so we can comment on them?"

That's a thought. It's time for a new posting anyway. Okay, let me see if I can think of anything else to add to the list and I'll start a new article.

Ho Ho said...

giant
"That was a 3.33Ghz Yorkfield CPU."

Actually that OC'd one was most certainly not Yorkfield. That demo was made before Yorkfield one. Unfortunately I can't find any links about it.


scientia
"It was running an aluminium HSF with a few heat pipes and a bog stock fan"

I'm sorry but how do we know if those heatpipes weren't attatched to a copper base? Because the fins are aluminum? There are loads of such heatsinks out there with copper base, heatpipes and aluminum fins. Is there any heatpipe coolers with aluminum base? Also copper vs aluminum fins doesn't make too much difference when there are heatpipes that carry the heat around. Copper is better at conducting heat, aluminum is better at disposing of it.

Btw, have you seen any comparisons of high-end aircoolers vs stock ones at removing heat from extremely hot CPUs? You might be suprised but often the stock ones can be almost as good as those highest-end ones, just they have insanely fast spinning fans that make them loud. So one could say that those high-end ones just do the same job as inbox only quieter.


"You can see that the fins are not copper, and other photographs confirm this."

I haven't once said they were.


"So, apparently your eroneous assumption is that this heatsink is comparable to something like a Zalman. It's not even close."

I'm still quite sure it is a lot better than any normal inbox cooler.


"I can safely say that that setup is not dumping 150 watts. "

Have you seen the coolers on 90nm P4D's? Those are good enough to cool those CPUs down and they are a lot worse than what is on that K10.

Unknown said...

Have you seen the coolers on 90nm P4D's? Those are good enough to cool those CPUs down and they are a lot worse than what is on that K10.

Exactly. The standard LGA775 fan came with the old Pentium Extreme Edition 840. That thing used more power than anything before it in desktop CPUs.

Unknown said...

will no one bother to mention that AMD's SOI process does not respond to overvolting that effectively?

I mean, you can't just push the voltage through the roof on an AMD processor and get results like you would on an Intel.

Plus, no one said the fan was blaring, and if I remember correctly, the fan shown was the stock boxed opti fan, which didn't run very fast, and could only handle basic overclocks (though very quietly).

Azmount Aryl said...

That heatsink that we saw on 3.0GHz Barcy has a copper base. It is one of the standard fans that AMD ships with their CPUs. Some of the Opterons for 939 and AM2 came with one of those plus all the 6000+ models and some of the lower once.

I've dealt with those things many times so i couldn't not to recognize one, however i do find it rather sad that not one of so called IT enthusiast on this blog apparently haven't came around one. You haven't purchased any AMD CPUs lately haven't you? -_'

Scientia from AMDZone said...

ho ho

Again, this is false. That cooling setup is not better than stock. The heatpipes in my notebook can usually cool the processor passively. Then when power consumption picks up the fan on the bottom kicks on and blows out warm air.

The amount of heat that that 3.0Ghz Phenom setup can dump is limited by the stock fan and the cooling area of the fins. My notebook has at least that much fin area. The stock fan would be a little better than my notebook fan but not nearly enough to push 150+ watts. You would have a point if they had used a larger, higher velocity fan.

Axel said...

I've heard it said that Barcelona will have a one quarter headstart over Penryn. In fact, aside from the slowest Barcys it would appear otherwise.

If you click on the "Boxed Server/Workstation Processors" link, you'll find that Intel are apparently promising volume Harpertown availability to resellers on November 11, from 2.0 GHz to 3.16 GHz.

This is quite unfortunate for AMD.

Ho Ho said...

scientia
"That cooling setup is not better than stock"

It certainly is considerably better than this one.


"The amount of heat that that 3.0Ghz Phenom setup can dump is limited by the stock fan and the cooling area of the fins"

Fan air moving capability and generated pressure also have a lot to do with it.


"My notebook has at least that much fin area"

It must have as it has a lot worse fan than K10 inbox. Also if you compare it with that P4D cooler then you can see your notebook and K10 both have quite a bit more fin area, not to mention that heatpipes carry heat more efficiently than a block of solid copper and aluminum.


"The stock fan would be a little better than my notebook fan but not nearly enough to push 150+ watts"

How does Intel manage it with inferior cooling? People even OC their CPUs with that inbox and do it quite well. Sure, the CPU runs hot as hell but it works. I've even had one 3GHz 90nm singlecore P4 running under full load OC'd to 3.3GHz for 24h having 1mm air clearance between heatsink base and the IHS. The CPU was constantly throttleing at 100C+ but no application was unstable.


"You would have a point if they had used a larger, higher velocity fan."

From the looks of it that was 80mm fan on the K10. Intel Inbox should have a bit smaller (7.xx IIRC). Btw, does anyone have any idea at what RPM were those fans moving?

Also one thing to know is that we could keep on arguing here for days but all that is quite meaningless if we don't know what was the actual CPU tempperature. The higher it was the more efficient cooling would become. It is basic physics: bigger temperature difference leads to faster heat transfer. Had this CPU ran at 90C+ that heatsink would be able to remove a lot more heat than from a 60C CPU.

abinstein said...

Ho Ho -
"3GHz with heatpipe cooler. Scientia calls those "watercooler-equivalent" and sais that shows pretty much nothing about what speeds could CPUs be released :)"

Please, that cooler is so small in size that it doesn't matter whether it uses copper or has heatpipe. The fact that it's using copper heatpipe shows that the cooler is aluminum-based with a copper core; otherwise it'd need no "copper" heatpipe, but just the whole copper trunk and fans.

This puny small cooler is obviously not comparable to the huge cooler used by Core 2 overclockers. Somehow your insistence on doubting this obvious fact makes me put more question marks on your ....

I guess I better stop or scientia will warn/censor me again. Still it's quite a fact that you're just picking because you can't swallow the facts.

Ho Ho said...

abinstein
"Please, that cooler is so small in size that it doesn't matter whether it uses copper or has heatpipe."

Small compared to what? I must say it is one of the best inbox coolers I've seen, though I haven't seen the solution that cools down those 150W+ MP Netbursts and >170W Powers.


"This puny small cooler is obviously not comparable to the huge cooler used by Core 2 overclockers"

Depending on who are you comparing with. Lots of people use inbox coolers for modest 20-30% OC with Core2, especially for the low-end stuff. Sure, temperature might get a bit high but it won't make the CPU not working.

Also stuff like Freezer series doesn't really cool much better than inbox, they just do it quieter.

Did everyone really miss that little smily I had there? My point was it is over the average inbox cooler, I wasn't saying it is comparable to Zalman 9000 series or something like that.


In totally unrelated news, Intel gets some new G stepping Xeons out with much lower idle power usage.

abinstein said...

The fact that it's using copper heatpipe shows that the cooler is aluminum-based with a copper core; otherwise it'd need no "copper" heatpipe, but just the whole copper trunk and fans.

Oops there's a typo. What I meant is the whole copper trunk and fins. The fan is small enough, too, though.

Ho Ho -
"It certainly is considerably better than this one."

Stop fooling around with things you don't understand, will you? A small cooler with a few copper heatpipes is not better than a 2x bigger copper-core cooler with extra aluminum fins and a powerful fan.


"It must have as it has a lot worse fan than K10 inbox. Also if you compare it with that P4D cooler then you can see your notebook and K10 both have quite a bit more fin area,"

What fin area do you see from that Phenom picture? My Turion-based notebook has one single heatpipe and fin area. Are you nuts?


"not to mention that heatpipes carry heat more efficiently than a block of solid copper and aluminum."

No, you are wrong. Heatpipe is effective in heat transfer, not dispersion. It transfers heat from a limited space to somewhere else with large surface area. Do you see any large surface area on the Phenom demo? You must be living in your imagination if you do.

abinstein said...

Ho Ho -
"In totally unrelated news, Intel gets some new G stepping Xeons out with much lower idle power usage."

Another indication that Intel is following AMD.

Aguia said...

alex,

Intel already said 45nm will be another paper launch.
2% max of all Intel production by year end.
Ramped production will happen in Q3/2008, guess what? We are in Q3/2007. You still have to wait 1 year to "really" see anything.
But at least they can brag that they are at 45nm when the others are not.

PS: Almost all of the production will be 65nm parts but like ho ho says publicity/advertising does it all, and Intel automatically only produces 45nm parts and all processors are Penryn’s.
When Conroe was released accounted “only” 5% but it seamed that all Intel processors were Core 2 Duo and Pentium 4 and Pentium D didn’t exist no more, immediately vanishing from the entire web from all the reviews…

Aguia said...

ho ho,
AMD is already years ahead of Intel in that department. In fact the Anandtech review was so ridiculous showing how good power now technology is (seamed that it was a new feature in AMD CPUs) when that technology already exists in AMD processors since 2003.
I wonder if that got removed/disable from previous reviews to make Intel not look so bad when they had the Netburst Xeons...

PS: Just look at your own posted review.
One 3800+ at 2.0Ghz 1MB L2 90 nm VS One OC 4.0Ghz 4MB L2 65nm.
The 3800+ wins tests with half the clock speed!!!!

According to our tests, Presler based CPUs working at 4.0GHz cannot outperform Athlon 64 FX-60 working at 2.6GHz only.

The 3800+ was consuming 65W full load (11W idle) while the OC PD was around 200W full load (65W idle). And those where the 65nm I can’t think what would happen with the 90nm Intel parts.

The K8 never looked that bad VS Conroe, at best Conroe as 600Mhz clock advantage (AMD needs 3.0Ghz to beat Intel 2.4Ghz) but that only in desktop applications, server doesn’t happen.

Ho Ho said...

abinstein
"Another indication that Intel is following AMD."

I guess Intel simply didn't thought about addint mobile CPU power states to server CPUs :)


axel
"2% max of all Intel production by year end."

If Scientias number for AMD is correct (10% of all are Opterons) then 2% of all Intel CPUs is quite a lot considering those CPUs will be on market for less than two months vs AMD nearly four months.


"Ramped production will happen in Q3/2008"

Seruious ramping begins sometime in Q1. In Q2 Intel intends to have lots of its notebooks on 45nm (40% IIRC) and that alone should be nearly 25% of all Intel desktop+notebook CPUs.


The 3800+ wins tests with half the clock speed!!!!"

It only won in FarCry, the rest were won by FX60. Also that P4 wasn't really well optimized as 4GHz was often slower than 3.9GHz one.

Aguia said...

I guess Intel simply didn't thought about addint mobile CPU power states to server CPUs :)

Well Intel learned their lesson when people start buying Notebook with Pentium 4 inside. The Pentium III wasn’t selling in mobile and Intel smartly renamed Pentium III as Pentium M and released the Centrino brand marketing campaign.
Notebook maker still keep the Pentium 4 in mobile, but started to disappear because people wanted Centrino.

And they are doing that only on server because they know OEMs won’t do notebooks with server sockets for sure, with the desktops they would lose the sales and the premium profits that come from mobile CPUs.

abinstein said...

Ho Ho -
"I guess Intel simply didn't thought about addint mobile CPU power states to server CPUs :)"

Save your smileys and just commit that you were fuding about Phenom's heatsinks. You have been a fuder and nothing more, on this matter and a few others.

I have warned you to stay on your own field of knowledge and do not mess around things that you don't understand. You have refused to listen and kept fuding. It's crystal clear now: 1) you're a Intel fanboys and fabricate biased information, 2) most anything you say other than graphics is as good as bullock.

Unknown said...



Intel already said 45nm will be another paper launch.


Show me a link with someone from Intel saying this. Just because 45nm CPUs will be limited in quantity this year doesn't mean it's a paper launch. I understand that K10 stock will be limited to two models on launch... and those will be limited in quantity; that must make it a paper launch!

When Conroe was released accounted “only” 5% but it seamed that all Intel processors were Core 2 Duo and Pentium 4 and Pentium D didn’t exist no more, immediately vanishing from the entire web from all the reviews…

So when AMD releases K10 and it accounts for only a small percentage of it's CPU sales do you think they will keep hyping K8 Opterons and getting sites to review K8 CPUs? Or could it be that AMD's marketing efforts will focus on K10 and the improvements it offers while they slowly phase out K8?

Ho Ho said...

abinstein
"Save your smileys and just commit that you were fuding about Phenom's heatsinks."

Please list incorrect things and FUD.

Hyping K10 will be a bit difficult, especially with the new 3.2GHz models of X2 and Opteron.

Ho Ho said...

Sorry, missed one of your posts there

abinstein
"The fan is small enough, too, though."

How small is it in mm?


"A small cooler with a few copper heatpipes is not better than a 2x bigger copper-core cooler with extra aluminum fins and a powerful fan."

Then why does AMD use it on its highest-end CPUs? Do you have any information about its performance compared to any other heatink?


"What fin area do you see from that Phenom picture? My Turion-based notebook has one single heatpipe and fin area. Are you nuts?"

You should read more careful next time or you'll miss posts just as I did. Scientia told his notebook has similar fin area as that K10 cooler. You should ask him.


"Heatpipe is effective in heat transfer, not dispersion"

Isn't that exactly what I said there?


"Do you see any large surface area on the Phenom demo?"

There are a lot of thin fins there, a lot more than on P4 inbox.


Please consider those things while making your list.

abinstein said...

"How small is it in mm?"

Small enough to fit under and be covered by a standard 60x60mm fan. Hell, even my old Socket A cooler looks larger than it.


"Then why does AMD use it on its highest-end CPUs? Do you have any information about its performance compared to any other heatink?"

So you first claim Phenom must be very hot because it is cooled by a terrific cooler, then claim the cooler must be terrific because it is used to cool the must-be-very-hot Phenom? What's your logic or lack of it anyway?


"You should read more careful next time or you'll miss posts just as I did. Scientia told his notebook has similar fin area as that K10 cooler. You should ask him."

I don't know what else to call you except a FUDer. You are willing to make totally erroneous remark when you don't even understand what scientia meant. The fin area of a standard turion-based notebook is so small that it's just sufficient to cool a 31/35W TDP processor. The area is certainly less than any standard desktop cooler, which is designed to cool at least twice the heat. Does that mean a lot to you?


"Isn't that exactly what I said there?"

You were implying the mere existence of heatpipes mean superior cooling ability. I am reminding you that transfer and dispersion are two different things.


"There are a lot of thin fins there, a lot more than on P4 inbox."

Where do you see that? Your habit of making FUDs and talking about things that you don't know is truly astounding.

Ahmar Abbasi said...

I like how everyone here is soo confident about Phenom shown running at 3.0Ghz when even AMD itself has said that this is not releasing anytime soon and the only official thing they have stated is that they will release barcelona at 2.0GHz for rite now.

Yet we have all these AMDers speaking soo confidentally about phenom and barcelona yields you would think they were actually there developing it when in reality they have no clue...........

Aguia said...

Giant,
I understand that K10 stock will be limited to two models on launch... and those will be limited in quantity; that must make it a paper launch!

Yes it will be a paper launch for sure. Even worst than conroe if we account the number of models introduced.

I can’t find the real link but this is almost as good:
Intel to Launch 45nm Desktop Chips in Q4

So when AMD releases K10 and it accounts for only a small percentage of it's CPU sales do you think they will keep hyping K8 Opterons and getting sites to review K8 CPUs? Or could it be that AMD's marketing efforts will focus on K10 and the improvements it offers while they slowly phase out K8?

Well that already happens with GPU reviews. I think that if 7600GS and 7900GS are still selling in the market, they should keep those products in the reviews.
With CPUs the same. Not only this helps too show the market as it helps the "future" buyer: "Let me see how my old P4 2.8Ghz is doing against this Core 2 Duo and this Athlon X2 in this game/application”. Just one of that generation models would be enough, instead of having various different clock speed processor of the same generation.
Unfortunately bad reviews are all over the place.

Ho Ho said...

abinstein
"Small enough to fit under and be covered by a standard 60x60mm fan."

Are you sure?

Btw, that fan can run at up to 6000RPM, enough to cool down even the hottest CPUs.


"So you first claim Phenom must be very hot because it is cooled by a terrific cooler, then claim the cooler must be terrific because it is used to cool the must-be-very-hot Phenom?"

No, I just said it could be hot. They didn't show the temperature numbers and my logic tells me that CPU with such OC will be really hot. If you have some other theory I'd be glad to hear it.


"What's your logic or lack of it anyway?"

1) It is (one of) the best inbox coolers out there
2) that CPU is (massively) OC'd and needs good cooling.


" You are willing to make totally erroneous remark when you don't even understand what scientia meant."

He said:
"The amount of heat that that 3.0Ghz Phenom setup can dump is limited by the stock fan and the cooling area of the fins. My notebook has at least that much fin area."

Emhasis mine. What did I miss? Only thing I can think of is that perhaps Scientia made a mistake when he said that his notebook has as much fin area as that K10 heatsink.


"The fin area of a standard turion-based notebook is so small that it's just sufficient to cool a 31/35W TDP processor."

Fin area is small in notebooks because there isn't much room. Also I remind you yet again that fin area isn't the only thing that shows how much that thing can cool down. Air pressure and volume is more important than that.


"You were implying the mere existence of heatpipes mean superior cooling ability"

Yes, heatpipes do move heat more efficiently than solid copper/aluminum, not to mention thin fins. Don't you agree?


"I am reminding you that transfer and dispersion are two different things."

I said "heatpipes carry heat more efficiently than a block of solid copper and aluminum". From what I know "carry" means the same as "transfer" in English. I said nothing about heatpipe dispersion ability.


"Where do you see that?"

On the picture. P4 inbox heat sink has considerably fatter, shorter and fewer fins than that one in the Inq picture. Don't you agree?



Please stop putting words in my mouth. Also you should tone down a bit, I haven't seen anyone as rude on this blog for a long time. If you think I'm wrong then say so, preferrably with something that shows I'm wrong. So far you've failed to do that.



aguia
"Unfortunately bad reviews are all over the place."

It takes a lot of time and effort to benchmark video cards. Usually reviews are done in limited time and they don't have enough time to compare lots of different models. I know only THG graphs that show most GPUs, though there are rather few benchmarks.

abinstein said...

Ho Ho -
"Are you sure?"

So it's 70mm instead of 60mm. It's certainly not huge. The Pentium-D cooler, OTOH, looks much bigger than the chip package in the center square, and you claim it's not higher-end? You are fuding, aren't you?

"Btw, that fan can run at up to 6000RPM, enough to cool down even the hottest CPUs."

You are so much wrong. The high rpm is needed due to the small size. 6000rpm by 70x70mm is nothing compared to real high-end cooler, those for example come with the Pentium-D's.


"No, I just said it could be hot. They didn't show the temperature numbers and my logic tells me that CPU with such OC will be really hot."

Basically what you are saying is this: my "logic" says it's hot, and it could be hot, thus it's probably hot, and thus the cooler must be better than other ones that are 2x larger.


"1) It is (one of) the best inbox coolers out there
2) that CPU is (massively) OC'd and needs good cooling."


There is no logic here, or do you actually believe otherwise? The cooler is good, no dispute, but it's not better than anything that $30 can buy (or all those cooler makers can go out of business right away). Furthermore, how good or bad the cooler is has absolutely nothing to do with whether the processor is clocked above its rated frequency.


"Only thing I can think of is that perhaps Scientia made a mistake when he said that his notebook has as much fin area as that K10 heatsink."

There is no way a notebook cooler has as much fin area as any desktop cooler. But apparently you were claiming the opposite, whether scientia made a mistake or not. You were willing to take other's unproven statements and make FUD on it on and on. This is your problem, which you just don't see.


"Fin area is small in notebooks because there isn't much room. Also I remind you yet again that fin area isn't the only thing that shows how much that thing can cool down. Air pressure and volume is more important than that."

Actually it's air flow that is important. This is why large fans are more effective when running slower than small ones. This is why a 70x70mm fan with 6000rpm isn't nearly as effective as a 10x10cm fan with just 3000rpm. You do know this, don't you? Yet you FUD along.


"Yes, heatpipes do move heat more efficiently than solid copper/aluminum, not to mention thin fins. Don't you agree?"

Where do you get that "yes"? Wrong logic, again? Heatpipes move heat more effectively, yet they don't offer superior cooling by themselves unless the heat is moved to a larger area with higher air flow.

The whole necessity of the heatpipe in the AMD cooler is to move heat from the bottom copper core to the separate (& relatively small) aluminum fins. It's most likely a decision of cost. It's apparently cheaper to make copper heatpipes and aluminum fins separately and mechanically put them together, than either to make a whole copper-finned cooler or a copper-core soldered with aluminum fins.


"I said "heatpipes carry heat more efficiently than a block of solid copper and aluminum". From what I know "carry" means the same as "transfer" in English. I said nothing about heatpipe dispersion ability."

So in proper logic, your "carry heat more efficiently" doesn't mean "better cooling ability," especially when the cooler is small in size, because there is no better heat dispersion.


"On the picture. P4 inbox heat sink has considerably fatter, shorter and fewer fins than that one in the Inq picture. Don't you agree?"

Fatter fins have better heat conductivity; shorter fins are actually better for heat dispersion. More importantly, the P-D cooler is much larger, coupled with a larger fan.


"Please stop putting words in my mouth."

It seems to me that you not only like to FUD but also like to deny your own FUDs. Wonderful combination, isn't it?


"Also you should tone down a bit, I haven't seen anyone as rude on this blog for a long time."

I will tone down as soon as you stop FUDing, which unfortunately you simply don't.


"If you think I'm wrong then say so, preferrably with something that shows I'm wrong. So far you've failed to do that."

I've shown you FUDing multiple times (the 70mm fan cools better than 10+cm fans, turion cooler has lots of fin area, etc.). I've also pointed out your logic errors (Phenom could be hot so AMD is "massively" overclocking it? Or heatpipes have better cooling ability because it transfers but doesn't disperse heat?). You just have zero capability of learning better.

Ho Ho said...

abinstein
"The Pentium-D cooler, OTOH, looks much bigger"

Size isn't everything. linky
84mm fan, max speed 2600RPM. Quite slow one, isn't it?

Yes, it does have somewhat bigger fan but as I said it has less fin area and solid aluminum is not as good at transfering heat from the copper core to outer parts of the fins. Also the fan is considerably slower than AMD one. Even still that inferior Intel inbox is able to keep their hot Netburst at manageable temperatures. That AMD one could handle much hotter stuff than those.


"and you claim it's not higher-end?"

Yes, compared to AMD cooler it is not that great. I'd even say it is quite miserable even . As I said before that Intel one is still capable of cooling quite hot CPUs and as the link shows AMD one is able to cool much hotter ones. Heck, it beats even things like full-copper 3 heatpipe >0.5kg Zalman 9500 series cooler with 92mm/2600RPM fan and huge fin area running at full speed. Also AMD cooler is better than Gigabyte G-Power Pro Heatsink with 3200RPm 110mm fan mounted on aluminum fins attatched to the base with four heatpipes.

Can you now finally understand that without a question AMD cooler simply is a whole lot better one than even P4D coolers? I knew it by simply comparing the images and using basic logic and physics skills, you couldn't seem to understand it even after lots of discussion and data. I sure hope you will finally understand it.

I'd like you to explain in detail what made you think AMD cooler was inferior to Intel one so I could tell you what your mistake was.


"The high rpm is needed due to the small size"

So basically it runs faster to cool more heat? Regular 4.5k rpm fans wouldn't be good enough?


"Basically what you are saying is this: my "logic" says it's hot, and it could be hot, thus it's probably hot, and thus the cooler must be better than other ones that are 2x larger."

Something like that, yes, but you'll need to put back the additional information I wrote before. I could pick out words from your talk to claim pretty much anything too.


"There is no logic here, or do you actually believe otherwise?"

Is 3GHz K10 not massively OC'd? If not then it should surely be on market really soon. Though AMD itself hasn't talked about CPUs nearly that fast.


"The cooler is good, no dispute, but it's not better than anything that $30 can buy"

As I've said most of those $30+ coolers cool just as good as inbox ones but they do it with less noise. Just a couple of days ago I read some article on it and they found out that Intel inbox was only a few degrees behind some relatively huge $40 cooler. Big difference was in generated noise. Observe


"Furthermore, how good or bad the cooler is has absolutely nothing to do with whether the processor is clocked above its rated frequency"

Are you claiming that CPU was not OC'd over its rated frequency?
With bad cooler you can't OC as well. I got +10% by replacing inbox with full copper Zalman 7000 series with my old Netburst. An Athlon XP got +5% going to some $25 cooler. I got higher clocks because I didn't want to run the inbox ones at their maximum speeds.

Btw, that 7000 series one had the exact problem I've been talking about. The base was around 4-5 degrees hotter than the fins as copper wasn't good enough at transferring heat to the tips of fins. Today with 9500 series heatpipe cooler difference between the base and fins is much smaller, in the order of 1-2C.


"Actually it's air flow that is important."

It depends. Airflow in itself doesn't do much if it doesn't put enough pressure on the fins (==molecules hit fans, remove some heat). Witl slowly moving air less molecules collide with fans. It is the same as with driving fast, air resistance grows exponentially to speed. Thus when you have two different sized fans with equal airflow the smaller one will be more effective.

Of course air pressure is meaningful only when it is applied directly to heatsink. Having small and fast casefans in place of big and slow ones doesn't make any difference.


"This is why a 70x70mm fan with 6000rpm isn't nearly as effective as a 10x10cm fan with just 3000rpm"

I don't remember the exact formulaes but I'd say the difference between the two probably isn't too big, assuming we are not talking about casefans but CPU ones.



"You do know this, don't you? Yet you FUD along."

Do you about what I just said about air pressure? For some reason it seems as you don't.


"Where do you get that "yes"? Wrong logic, again?"

Do I really have to say everything twice before you understand? My point was that heatpipes move heat faster and more effectively than solid metal and thus give the cooler better cooling abilities by moving heat to the distant parts of the heatsink faster.


"Heatpipes move heat more effectively, yet they don't offer superior cooling by themselves unless the heat is moved to a larger area with higher air flow."

As I just said they move heat faster and keep the overall heatsink at similar temperature. Bigger surface area isn't the only thing that matters. With solid heatsinks the core is rather hot and fin ends cold. With heatpipes the heatpipes themselves and fins near the contact place will be hottest. As heatpipes carry heat farther they distribute the heat more evenly over the fins.


"The whole necessity of the heatpipe in the AMD cooler is to move heat from the bottom copper core to the separate (& relatively small) aluminum fins."

Also this makes it easier for the fan to blow air through the fins. With Intel-like cooler you blow it almost straight against motherboard, with AMD one there is air clearance.

Those fins actually have quite a lot of surface area, each individual one more than one in Intel cooler.


"It's apparently cheaper to make copper heatpipes and aluminum fins separately and mechanically put them together, than either to make a whole copper-finned cooler or a copper-core soldered with aluminum fins."

Unfortunately I don't have any data on it but I highly doubt it. Intel-like cooler should be considerably cheaper to manufacture, all you need to do is to mold a piece of aluminum and put a piece of solid copper in the hole and you are done. That is considerably less work than building heatpipes and attaching tens or hundreds of fins to them.



"So in proper logic, your "carry heat more efficiently" doesn't mean "better cooling ability," especially when the cooler is small in size, because there is no better heat dispersion."

After consulting with dictionary I understood that I have misunderstood the meaning of the word "disperse". I thought it meant giving heat directly to air but it seems as you can also use it in place of carry and transfer. I think I might need a bit more language practice, English is not my first language.


"It seems to me that you not only like to FUD but also like to deny your own FUDs."

What you call "denying of FUD" most people call discussion. Seems I just need to keep on reiterating myself and use simplier words or you might yet again misunderstand.


"I've shown you FUDing multiple times (the 70mm fan cools better than 10+cm fans"

As I said it could, it depends on the 10cm one.


"turion cooler has lots of fin area"

As I said, I only took what Scientia said. It did seem weird that his notebook has so big heatsink but who knows, those desktop-replacements might have bigger coolers.


"I've also pointed out your logic errors (Phenom could be hot so AMD is "massively" overclocking it?"

Given it has 1.9 or luckily, if to believe the delay rumours, 2.0GHz release speed I'd say that 3GHz really is massively OC'd. Didn't Scientia just have some article where he was talking about Intel doing somewhat similar demos with OC'd CPUs?


"I've shown you FUDing multiple times"

... and this isn't the first time I've proven you wrong. We both learn from each other from time to time :)

Next thing I suggest you to learn is writing without so many accusations of FUD'ing and other pointless stuff, you are wasting everyones time by doing it and it only makes your text look less credible. If you think something is wrong then say (prove) it, no need for extra words.

Ho Ho said...

Yet some more unrelated news. Later this week Intel is going to publis several research papers. Topics are as follows:


1. “Datacenter-on-a chip” (pure enterprise)
2. Physical modeling for realism in games/movies (both client and server)
3. Home multi-media search/mining (a client app)
4. Packaging & integration of this L4 cache
5. On-die integration of many cores.
6. A proposed hardware task scheduler
7. Giving an IA “look and feel” to the programming of integrated hardware accelerators
8. A comprehensive run-time environment for tera-scale platforms

Short introduction to the papers is here. Thre should be enough information for several lengthy and interesting articles. I hope they shed some light on their real-time ray tracing research also :)

abinstein said...

Ho Ho -
"Yes, compared to AMD cooler it is not that great. I'd even say it is quite miserable even ."

Nice comparison chart. However, it doesn't prove your point. The AM2 stock cooler performs better due to the high rpm, thus the high noise level. It is not due to the copper heat pipes or the thin aluminum fins. In fact, at "low" setting, the 3800rpm is still higher than the fans on most other heat sinks.


"As I said before that Intel one is still capable of cooling quite hot CPUs and as the link shows AMD one is able to cool much hotter ones."

All it says is the AMD cooler can work as a more capable cooler. Still your lack of logic is apparent when you translate this to the "conclusion" that 1) copper heatpipe offer superior cooling by itself, 2) the 3.0GHz Phenom is massively "overclocked", 3) the 3.0GHz Phenom must consume a lot of power. All above are FUDs.


"My point was that heatpipes move heat faster and more effectively than solid metal and thus give the cooler better cooling abilities by moving heat to the distant parts of the heatsink faster."

But what you said is false.. Heatpipe can transfer heat more efficient if there is good temperature gradient. If the heat is transferred to a set of small aluminum fins with less efficient dispersion (e.g., slow air flow), the temperature of both the fins and the heatpipes rises and its gradient reduces, resulting in less and less heat transfer.

The reason of AM2 cooler's good performance is not (just) copper heat pipes but more its high fan speed. At 3800rpm (even higher than the highest fan speed of P-D stock cooler), the performance of AM2 cooler is poorer than that of P-D stock.


"Given it has 1.9 or luckily, if to believe the delay rumours, 2.0GHz release speed I'd say that 3GHz really is massively OC'd."

You are mixing up releasable speed and attainable speed. A good quantity is required for the former, whereas a minimal is enough for the latter. Both however are not directly relevant to whether a chip is being "overclocked" or not.


"Didn't Scientia just have some article where he was talking about Intel doing somewhat similar demos with OC'd CPUs?"

Are you referring to the 3.3Ghz Penryn demo? I don't know what scientia thinks, but it's basically a proof of how good Intel's 45nm can be. It has as little indication with respect to releasable speed as the 3.0GHz Phenom has.

Whereas Penryn's 3.3GHz shows the potential of Intel's 45nm process, Phenom's 3.0GHz shows the potential of K10's design. So we know that with 45nm Intel not only reduces power but also increase speed. Similarly, we also know that with K10 AMD not only increases IPC but also attain the same clock speed.


"... and this isn't the first time I've proven you wrong. We both learn from each other from time to time :)"

True, I was wrong about the P-D stock cooler. It's a much worse & smaller one than I've expected. Still, most of your arguments are false (heatpipes, fins, etc.).

Axel said...

A while ago, Axel hypothesized:

It is likely that the 3.33 GHz Wolfdale dual cores will not exceed $266, as Intel evidently intends to exert pricing pressure to the extreme to ensure that AMD cannot make a profit on their Kuma dual cores.

and:

Intel will use their 45-nm die size advantage to their fullest to price AMD down to ASPs that cannot support their current CAPEX / R&D spend profile.

To which Scientia doubtfully replied:

Intel can't do that without cutting its own profits.

Well guess what, they're doing it. Intel are pricing Harpertown so low that I can't see even the fastest Wolfdale being over $266. Remember that Harpertown is the server quad-core and therefore would carry a premium over an identically clocked Yorkfield, the desktop quad-core. With a 2.33 GHz Harpertown selling for $256, a 2.33 GHz Yorkfield would likely be $216. So where would that put a 2.33 GHz Wolfdale? Around $120?

AMD are finished. Intel are rushing to push quad-core into the mainstream in order to force AMD to use more Fab 36 capacity for Barcelona instead of Kuma. Well, thanks to the 283 mm2 die and lower yields from monolithic design, AMD cannot afford to sell Barcys as cheap as Intel are offering Harpertown.

Kuma will be no help for ASPs, as it will be competing with ridiculously cheap and fast Wolfdales equipped with SSE4. As Intel have been evangelizing SSE4 to developers since April 2006, it is likely that many common synthetic and audio/video benchmarks will have been re-compiled for the new instruction set in time for Penryn's launch, showing impressive performance gains for many benchmarks. This flurry of marketing is sure to spoil AMD's launch of faster Barcys in Q4 07 and Q1 08.

Once Penryn ramps, it's game over for AMD unless K10 truly has a substantial per-clock or per-watt advantage. So far we have seen no indication of this beyond AMD Powerpoints. We have, however, seen a couple indications to the contrary but I know everyone's tired of hearing about Cinebench and POVray so let's just wait another month and get the whole scoop.

abinstein said...

"Well guess what, they're doing it. Intel are pricing Harpertown so low that I can't see even the fastest Wolfdale being over $266."

Suppose Wolfdale is priced 60% of Harpertown at the same clock and 2.33GHz Harpertown ~$250, then 2.33GHz Wolfdale will be around $150. Right now E6850 at 3GHz (~$320) is priced about 1.8x that of E6550 at 2.33GHz (~$180). Thus the fastest Wolfdale, if being 3.3GHz, should be priced about 2x the 2.33GHz Wolfdale, or roughly $300.

"Kuma will be no help for ASPs, as it will be competing with ridiculously cheap and fast Wolfdales equipped with SSE4."

I find your argument ridiculous. Intel claimed 45nm will represent small volume of Q4 (<15%?), and only reach cross over in Q2/Q3 next year. So basically what you say is the 10% Penryn-based processors will be so cheap, and their SSE4 instructions are so powerful, that they will destroy the 90% sales of Conroe-based processors.

Use your brain, is it possible?

BTW, AMD already said that Kuma will be very small volume and will have almost no impact on Q4 ASP. It will only contribute to ASP in Q1 2008 and later. IMO, K10 doesn't seem to be too competive in terms of enthusiast performance, anyway. However, for server, K10 does seem somewhat superior.

"As Intel have been evangelizing SSE4 to developers since April 2006, it is likely that many common synthetic and audio/video benchmarks will have been re-compiled for the new instruction set in time for Penryn's launch, showing impressive performance gains for many benchmarks."

No matter how many benchmarks is optimized for SSE4, the reality is that SSE4 helps almost nothing but video encoding, which can be done much faster with CTM anyway.

Pop Catalin Sever said...

Intel is lowering CPU prices to attack AMD because it can compensate with chipset prices on the server side. Intel server chipsets are overpriced.

Unfortunately lower CPU prices won't be enough in server space, AMD still can reach a large part of the market because on NUMA that allows higher cpu density per server, higher memory bandwidth, lower power consumption because they don't use FBDIM's, and the
performance even with current Opterons is comparable to Intel in server space not compared AMD core vs Intel core but platform vs platform with comparable density, price, power consumption (system power consumption and not CPU power consumption I feel I must repeat myself, CPU + Memory + Chipset).

Overall I think the AMD's server platforms are better balanced and with more value than Intel's with few notable exceptions.

Axel said...

Abinstein

So basically what you say is the 10% Penryn-based processors will be so cheap, and their SSE4 instructions are so powerful, that they will destroy the 90% sales of Conroe-based processors.

I already showed you how Intel have cannibalized their own 1066 FSB Conroes with the new 1333 FSB Conroes, but apparently you didn't pay attention. Prices from Newegg:

E6400 (2133/1066) - $205
E6550 (2333/1333 - $179

E6600 (2400/1066) - $230
E6750 (2666/1333) - $212

E6700 (2666/1066) - $317
E6850 (3000/1333) - $330

Note that the E6850 price is inflated due to excessive demand. Tray price is $266 and once demand settles down, Newegg's price will come down to the expected $280 or so.

So with this pretty solid proof that Intel are cannibalizing their own one-year old C2Ds with the 1333 FSB refresh, what makes you think they won't do it all over again with Wolfdale/Yorkfield in Q1 2008? Time to wake up and see what Intel is doing: strangling AMD with cutthroat pricing. This intention is obvious in their every action now.

Cheap Q6600s, cheap Harpertowns, cheap Wolfdales. Carpet bomb AMD until they're forced to outsource. Already there are rumors on Fudzilla that some 200-mm equipment in Fab 30 has been sold. I won't bother linking to that site because of how much they get wrong, but this particular rumor does have a ring of truth to it. Basically it looks like AMD intend to eventually sell off Fab 30, though right now their official position is that the Fab 38 conversion has simply been "delayed". In truth, their cash position is so desperate that Fab 30 is likely to be gone within a year. That's precisely Intel's intention: force AMD to give up fab capacity.

Unknown said...

I love how people like Axel enthusiastically and excitedly proclaim the death of AMD, without realizing that the death of AMD will result in what will essentially be the death of Intel.

With only one player in total in the entire personal computer processor market and the timely lawsuit from AMD, the FTC will very likely consider splitting up Intel. Sun, IBM, Via, and other processor manufacturers essentially cannot compete with Intel, not due to any deficiency in product quality, but due to the nature of this indaustry. They either do not have the liscensing rights to x86, do not have any experience with the process, or do not make most of their money off of it.

Someone more familiar with monopoly law could probably correct any misconceptions I have and give more detail, but basically we and Intel are all fubar'd if AMD folds, as this will be far different than M$'s situation as a monopoly.

And no, I'm not stating this as a reason that AMD will not die or will do better in the future, as I'm not certain or making a case for either. However, I am pointing out why AMD's misfortunes are something that no one should be happy about.

Aguia said...

Axel,
The price is down because they (Intel) know:
- 1333Mhz FSB motherboards cost more, especially if equipped with DDR3.
- Intel should stop (if didn’t already) production of 1066 FSB processors soon, so its prices has to be higher or there would be shortages of the “old” stuff.
- Intel knows it easier to OC one 1066Mhjz FSB processor then one 1333Mhz FSB, so they know Overclockers will obvious buy the 1066MHz version.

Resuming Intel has to give reasons to buy higher FSB processors otherwise they would not sell very well.


Cheap Q6600s, cheap Harpertowns, cheap Wolfdales.

Except Intel there are not other quad core processors. How do you know one Q6600 is cheap?

FAB30 will be upgraded from 200mm to 300mm. Why should AMD keep with the 200mm tools if they are no longer needed?

Aguia said...

I love how people like Axel enthusiastically and excitedly proclaim the death of AMD, without realizing that the death of AMD will result in what will essentially be the death of Intel.

The death of AMD is the death of Intel. The one that would stand still is the future Wii of the PC world, VIA.


but basically we and Intel are all fubar'd if AMD folds, as this will be far different than M$'s situation as a monopoly.

Very good point greg, after all there are no Linux, no Open Source, no freeware similar in the CPU world.
And we already have bad scenarios from NVIDIA. How many months have passed since the release of the 8800 series NVIDIA cards, the price doesn’t change?

Axel said...

Greg
I love how people like Axel enthusiastically and excitedly proclaim the death of AMD, without realizing that the death of AMD will result in what will essentially be the death of Intel.

I get excited about logic and truth, not the death of a company. The bottom line is there are big changes coming at AMD and unfortunately K10 isn't one of them.

abinstein said...

Axel -
"I already showed you how Intel have cannibalized their own 1066 FSB Conroes with the new 1333 FSB Conroes, but apparently you didn't pay attention."

Nop, I didn't pay attention because a glance at the evidence show that you have no clue. Intel priced 1333 Conroe low because
1) they require more expensive motherboards,
2) they consume more power,
3) most importantly, they do not run faster.


"Note that the E6850 price is inflated due to excessive demand. Tray price is $266 and once demand settles down, Newegg's price will come down to the expected $280 or so."

So your believe once 3.3GHz Wolfdale is released, its tray price will be less than $266, leaving 90% Conroe-based processors sold below what, $200? Right now at newegg more than 1/3 Core 2 Duo are sold above $200, and you expect after Wolfdale is out, the ratio will drop to 1/10, and it won't negatively affect Intel's ASP?

Do you understand that 10% sales volume simply cannot cannibalize the sales of the 90% others, but can happily push down ASP globally?


"The bottom line is there are big changes coming at AMD and unfortunately K10 isn't one of them."

K10 is not the only big changes, but certainly one of them.

abinstein said...

Oh well, I was probably wrong to have mixed up tray price and sales price. It seems after Intel's July price cut all its processors except E6850 are sold below $200.

According to digitimes, there are 4 known models of Wolfdale: 3.16GHz, 2.83GHz, 2.66GHz, and 2GHz. Given Wolfdale performs ~7% better than Conroe at the same clock, mostly due to the larger cache, I'd expect its top 2 models to be priced above current E6850. In other words, 50% of Wolfdale will replace current Conroe from E6850 down to E6550).

50% Wolfdale is less than 5% of Intel's Q4 volume. In other words, if 3.16GHz Wolfdale is priced at $266, then only less than 5% of Intel's dual-core will be priced above $160. If this is true, it's bound to affect Intel's ASP.

Axel said...

Abinstein

It seems after Intel's July price cut all its processors except E6850 are sold below $200.

You mean all its dual-core processors. Quad-cores range from $266 (Q6600) to ~$1300 (QX6850).

Now I don't think it's a big stretch of the imagination to suppose that when Yorkfield 3.33 GHz is launched in Q1 08, Q6600's tray price will come down from $266 to around $200. That pretty much means all dual-cores, except perhaps the very fastest Wolfdale, would be under $200.

K10 is not the only big changes, but certainly one of them.

You hope. And so do I, for the sake of exciting times ahead and good competition. But I don't have much faith that AMD will really deliver with K10. It might do fine in the server space, but it's the other markets that bring in the revenue. I believe that K10 will not, on its own merits, bring AMD back to profitability. It's not good enough. For AMD to survive they have to make a fundamental change to their operations. Outsource, go private, etc. If they continue as they're going, it's only a couple quarters to BK.

abinstein said...

axel -
"You mean all its dual-core processors. Quad-cores range from $266 (Q6600) to ~$1300 (QX6850)."

QX6850 is sold at $999. Quad-core is very small percentage in 2007 desktop, anyway. With the introduction of Yorkfield this is expected to change. However, AMD's K10 are actually more competitive as quad-cores.


"Now I don't think it's a big stretch of the imagination to suppose that when Yorkfield 3.33 GHz is launched in Q1 08, Q6600's tray price will come down from $266 to around $200."

Nop, again according to digitimes, the fastest Yorkfield to be released in Q4'07 is just shy of 3GHz. Q1 or maybe Q2'08 will come the 3.16GHz, with 3.33GHz in the uncertain future.

On the dual-core side, fastest Wolfdale will be 3.16GHz in Q4'07, and move to 3.33GHz in Q1 or Q2'08.

abinstein said...

Axel -

There are FUDers and bad FUDers; the latter not only FUD but also do so with no sense. Unfortunately you are the latter.


"You hope. And so do I, for the sake of exciting times ahead and good competition."

You don't need to hope, just observe. Right at the dawn of Phenom, Intel is making price cuts one after another on all of its processors, from quad-core to dual-core. Apparently Intel is afraid of Phenom more than you do.


"But I don't have much faith that AMD will really deliver with K10. It might do fine in the server space, but it's the other markets that bring in the revenue."

So basically you admit K10's design superiority, just don't think AMD can deliver K10 with large enough quantity for the desktop? Actually, with Phenom, AMD only needs to take more share from the top spot (quad-core). You will see AMD pushing multi-core more aggressively than Intel starting Q4 2007.


"I believe that K10 will not, on its own merits, bring AMD back to profitability."

No, not even K8 was enough to bring AMD back to profitability. The question is not whether AMD will be as profitable as Intel, or whether AMD's stock price will rise as high as 2006 - those are for the investors concern; the question is whether AMD will grow stronger and bring more competition to Intel, both of which benefit their customers, like you and me. The answer, with the introduction of K10, is certainly yes.


"For AMD to survive they have to make a fundamental change to their operations. Outsource, go private, etc. If they continue as they're going, it's only a couple quarters to BK."

You sound as ridiculous as Sharikou now. Just because Intel had a few bad quarters in 2005 he claimed the company is going "BK" (did he even invent that term?); likely, just because AMD had a fe bad quarters in 2007 you are claiming it is going "BK" too.

Even worse, your "solutions" are ridiculous, too. How does going private help AMD? The challenge the company facing is not being public of private, but lack of high-end (CPU) product and low ASP. Going private only makes things worse since changing hand always tend to interrupt or delay R&D.

AMD is outsourcing many stuff, but certainly not the processor manufacturing. The fact is there is no other place on earth that can produce K8/K10 as efficiently as AMD's FAB30/FAB36.

Amdzoner said...

Albstein, you have some data wrong.

It's Harpertown for 2P that will launch up to 3,16Ghz on the server side on November 11th.

Yorkfield Extreme Edition is setting for a launch date in December, probably late. This is a 3,33Ghz.

abinstein said...

"It's Harpertown for 2P that will launch up to 3,16Ghz on the server side on November 11th."

I have no data regarding Intel's 45nm server clock speed, but I find it improbable to go above 3GHz. It's only 11% of Intel's server share, anyway.

"Yorkfield Extreme Edition is setting for a launch date in December, probably late. This is a 3,33Ghz."

Extreme Edition is by definition placed outside the regular pricing structure. It will be most likely priced above $1000.

Mo said...

Hey Sci,

I have an idea for you, you can also make a few extra bucks too.

Why don't you go write for Forbes and Wall Street Journal? I mean, obviously all these analysts with their years of experience/MBA's are WRONG.
You obviously state that BLUNTLY in your posts at AMDZone.

Is EVERY person who says something negative about AMD paid by Intel? Doesn't this statement get ridiculous?

thoughts? Yes This in ref to your response to J. Pitzer.

Axel said...

Time to resolve some FUD.

Abinstein

Nop, I didn't pay attention because a glance at the evidence show that you have no clue. Intel priced 1333 Conroe low because
1) they require more expensive motherboards,
2) they consume more power,
3) most importantly, they do not run faster.


Lol, now you're spouting nonsense. Let's take each point one by one:

1) No they don't. One can run the new 1333 FSB Conroes just fine on many P965 boards. I myself have been running a P965 board at 1600 FSB for over a year now.

Second, OEMs will be soon making all their boards 1333 FSB capable soon anyway. Once they do, why on earth would they buy E6400s from Intel when they can get E6550s for less that run 200 MHz faster and also have double the L2 cache?? Again, look again at what you're clearly missing:

E6400 (2133/1066/2MB) - $205
E6550 (2333/1333/4MB) - $179

2) No. The new CPUs are G0 stepping, which consume less power per clock than the older generation.

3) Again, look at the E6400 vs. E6550 above. For less money the OEMs are getting a Conroe that is 200 MHz faster, runs on a faster FSB, and has double the L2 cache.

Bottom line, you're throwing out random nonsense because you still refuse to believe that Intel are doing this specifically to bury AMD.

50% Wolfdale is less than 5% of Intel's Q4 volume. In other words, if 3.16GHz Wolfdale is priced at $266, then only less than 5% of Intel's dual-core will be priced above $160. If this is true, it's bound to affect Intel's ASP.

Of course it will affect Intel's ASP, for dual-core. However, a greater mix of quad-core might balance this out. Anyway, you are once again refusing to understand Intel's willingness to sacrifice ASPs and profits in the near-term to cripple AMD and win long-term.

Quad-core is very small percentage in 2007 desktop, anyway. With the introduction of Yorkfield this is expected to change.

No, Intel's guidance on quad-core adoption in 2007 is sure to change in their next earnings call. Already 204 reviews of Q6600 on Newegg, these are selling like mad.

And HP are selling two PCs equipped with Q6600. Dell don't seem to have any yet.

Right at the dawn of Phenom, Intel is making price cuts one after another on all of its processors, from quad-core to dual-core. Apparently Intel is afraid of Phenom more than you do.

Lol, Intel made the price cuts on July 22, at least five months before Phenom is due. Clearly the cuts had nothing to do with Phenom. Again, you're not acknowledging Intel's true motive here, which is to finish AMD.

So basically you admit K10's design superiority, just don't think AMD can deliver K10 with large enough quantity for the desktop?

Thanks for putting words in my mouth, I said no such thing. In fact, I've said quite the opposite for some time now. I believe that K10 will lose to Penryn per clock in the desktop, workstation, and mobile markets due to Penryn's faster SSE performance, parity in integer performance, along with adequate memory bandwidth for those spaces.

Now K10's superior memory bandwidth should give it an advantage over Penryn in some server applications. Overall in the server space I think Barcy and Harpertown will be pretty close in both performance per-clock and per-watt. But Harpertown's MCM approach and overall smaller die will give it the big advantage in pricing.

Nop, again according to digitimes, the fastest Yorkfield to be released in Q4'07 is just shy of 3GHz. Q1 or maybe Q2'08 will come the 3.16GHz, with 3.33GHz in the uncertain future.

On the dual-core side, fastest Wolfdale will be 3.16GHz in Q4'07, and move to 3.33GHz in Q1 or Q2'08.


It would have been nice if you'd linked to it. Anyway, for both Wolfdale and Yorkfield one of the speed bins is unspecified. I bet this will be the wildcard that Intel will play depending on Barcelona performance and market uptake. I'm sure they can push a 3.33 GHz Wolfdale and 3.16 GHz Yorkfield out on initial launch if they have to.

The question is not whether AMD will be as profitable as Intel, or whether AMD's stock price will rise as high as 2006 - those are for the investors concern; the question is whether AMD will grow stronger and bring more competition to Intel, both of which benefit their customers, like you and me.

No. If they continue to burn cash at several hundred million dollars per quarter, there is a very real possibility that AMD will end up like Cyrix in 2008.

I have no data regarding Intel's 45nm server clock speed,

Yes you do.

Aguia said...

Another good article about K10.

AMD K10 Micro-Architecture

Scientia from AMDZone said...

Axel

"A while ago, Axel hypothesized:

It is likely that the 3.33 GHz Wolfdale dual cores will not exceed $266,

Intel are pricing Harpertown so low that I can't see even the fastest Wolfdale being over $266."


I'm trying to figure out what your point is. The prices you listed kick in in November. This means that will see the desktop prices drop in Q1 08. It makes perfect sense to me that dual core prices will drop as quad core ramps. Are you saying something different from this?

"AMD are finished. Intel are rushing to push quad-core into the mainstream in order to force AMD to use more Fab 36 capacity for Barcelona instead of Kuma."

That was already the plan. Most of the production in Q4 will be server chips. I'll leave the rest of this for the next article which includes ASP projections.

Scientia from AMDZone said...

mo

You didn't read very carefully. I said that his rating was correct but for the wrong reason. K10 is not enough volume in 2007 to have much affect on AMD's profitability. What would this have to do with performance, clock speeds, or support? And, his statement about support obviously ludicrous. K10 drops into the same AM2 and F sockets as current K8's so what could possibly be lacking in support? If you feel that your common sense is worse than his "analysis" then by all means believe him.

Scientia from AMDZone said...

Oh, before I forget. Having seen azmount's picture I would say the heatpipe in my notebook has about half of that amount of fin area. Of course, it is all copper though instead of aluminum.

Scientia from AMDZone said...

axel

I think your ideas about bankruptcy are way off the mark. However, I'll have to agree with you about 1333Mhz which should be accepted rapidly and not increase price. Likewise, the G0 stepping is very good. That is the first Intel 65nm stepping that I feel is capable of clocking to 3.2Ghz.