Monday, November 05, 2007

Has Intel's Process Tech Put Them Leagues Ahead?

There has been a lot of talk lately suggesting that Intel is way ahead of AMD because of superior R&D procedures. Some of the ideas involved are rather intriguing so it's probably worth taking a closer look.

You have to take things with a grain of salt. For example, there are people who insist that it wouldn't matter if AMD went bankrupt because Intel would do its very best to provide fast and inexpensive chips even without competition. Yet these same people will, in the next breath, also insist that the reason why Intel didn't release 3.2Ghz chips in 2006 (or 2007) was because, "they didn't have to". I don't think I need to go any further into what an odd contradiction of logic that is. At any rate, the theory being thrown around these days is that Intel saw the error of its ways when it ran into trouble with Prescott. Then it scrambled and made sweeping changes to its R&D department, purportedly mandating Restricted Design Rules so that the Design staff stayed within the limitations of the Process staff. The theory is that this has allowed Intel to be more consistent with design and to leap ahead of AMD which presumably has not instituted RDR. The theory also is that AMD's Continuous Transistor Improvement has changed from a benefit to a drawback. The idea is that rather than continuous changes allowing AMD to advance, these changes only produce chaos as each change spins off unexpected tool interactions that take months to fix.

The best analogy of RDR that I can think of is Group Code Recording and Run Length Limited recording. Let's look at magnetic media like tape or the surface of a floppy disk. Typically a '1' bit is recorded as a change in magnetic polarity while a '0' is no change. The problem is that this medium can only handle a certain density. If we try to pack too many transitions too closely together they will blend and the polarity change may not be strong enough to detect. Now, let's say that a given magnetic medium is able to handle 1,000 flux transistions per inch. If we record this directly then we can do 1,000 bits per inch. However, Frequency Modulation puts an encoding bit between data bits to ensure that we don't get two 1 bits in a row. This means that we can actually put 2,000 encoded bits per inch and of this 1,000 bits is actual data. We can see that although FM expanded the bits by a factor of 2 there was no actual change in data density. However, by using more complex encoding we can actually increase density. By using (1,7) RLL we can record the same 2,000 encoded bits per inch but we get 1,333 data bits. And, by using (2,7) RLL we space out the 1 bits even further and can double the recording density to 4,000 encoded bits per inch. This increases our data bits by 50% to 1,500. GCR is similar as it maps a group of bits into a larger group which allows elimination of bad bit patterns. You can see a detailed description of MFM, GCR, and RLL at Wikipedia. The important point is that although these encoding schemes initially make the data bits larger they actually allow greater recording densities.

RDR would be similar to these encoding schemes in that while it would initially make the design larger it would eliminate problem areas which would ultimately allow the design to be made smaller. Also, RDR would theoretically greatly reduce delays. When we see that Intel's gate length and cache memory cell size are both smaller than AMD's and we see the smooth transition to C2D and now Penryn we would be inclined to give credit to RDR much as EDN editor, Ron Wilson did. You'll need to know that OPC is Optical Proximity Correction and that DFM is Design For Manufacturability. One example of OPC is that you can't actually have square corners on a die mask so this is corrected by rounding the corners to a minimum radius. DFM just means that Intel tries very hard not to design something that it can't make. Now, DFM is a good idea since there are many historical examples of designs from Da Vinci's attempts to cast a large bronze horse to the Soviet N1 lunar rocket that failed because manufacturing was not up to design requirements. There are also numerous examples from the first attempts to lay the Transatlantic Telegraph Cable (nine year delay) to the Sidney Opera House (eight year delay) that floundered at high cost until manufacturing caught up to design.

I've read what both armchair and true experts have to say about IC manufacturing today and to be honest I still haven't been able to reach a conclusion about the Intel/RDR Leagues Ahead theory. The problems of comparing the manufacturing at AMD and Intel are numerous. For example, we have no idea how much is being spent on each side. We could set upper limits but there is no way to tell exactly how much and this does make a difference. For example, if the IBM/AMD process consortium are spending twice as much as Intel on process R&D then I would say that Intel is doing great. However, if Intel is spending twice as much then I'm not so sure. We also know that Intel has more design engineers and more R&D money than AMD does for the CPU design itself. It seems that this could be the reason for smaller gate size just as much as RDR. It is possible that differences between SOI and bulk silicon are factors as well. On the other hand, the fact that AMD only has one location (and currently just one FAB) to worry about surely gives them at least some advantage in process conversion and ramping. I don't really have an opinion as to whether AMD's use of SOI is good idea or a big mistake. However, I do think that the recent creation of the SOI Consortium with 19 members means that neither IBM nor AMD is likely to stop using SOI any sooner than 16nm which is beyond any current roadmap. I suppose it is possible that they see benefits (from Fully Depleted SOI perhaps) that are not general knowledge yet.

There is at least some suggestion in Yawei Jin's doctoral dissertation that SOI could have continuing benefits. The paper is rather technical but the important points are that SOI begins having problems at smaller scale.

"we found that even after optimization, the saturation drive current planar fully depleted SOI still can’t meet 2016 ITRS requirement. It is only 2/3 of ITRS requirement. The total gate capacitance is also more than twice of ITRS requirement. The intrinsic delay is more than triple of ITRS roadmap requirement. It means that ultra-thin body planar single-gate MOSFET is not a promising candidate for sub-10nm technology."

The results for planar double gates are similar: "we don’t think ultra-thin body single-gate structure or double-gate structure a good choice for sub-10nm logic device."

However, it appears that "non-planar double gate and non-planar triple-gate . . . are very promising to be the candidates of digital devices at small gate length." But, "in spite of the advantages, when the physical gate length scales down to be 9nm, these structures still can’t meet the ITRS requirements."

So, even though AMD and IBM have been working on non-planar, double gate FinFET technology, this does not appear sufficient. Apparently this would have to be combined with novel materials such as GaN in order to meet the requirements. It then appears that it is possible for AMD and IBM to continue using SOI down to a scale smaller than 22nn. So, it isn't clear that Intel has any longterm advantage by avoiding SOI based design.

However, even if AMD is competitive in the long run that would not prevent AMD from being seriously behind today. Certainly when we see reports that AMD will not get above 2.6Ghz in Q4 that sounds like anything but competitive. When we combine these limitations with glowing reports from reviewers who proclaim that Intel could do 4.0Ghz by the end of 2008 this disparity seems insurmountable. The only problem is that the same source that says that 2.6Ghz Phenom will be out in December or January also says Fastest Intel for 2008 is 3.2GHz quad core.

"Intel struggles to keep its Thermal Design Power (TDP) to 130W and its 3.2GHz QX9770 will be just a bit off that magical number. The planned TDP for QX9770 quad core with 12MB cache and FSB 1600 is 136W, and this is already considered high. According to the current Intel roadmap it doesn’t look like Intel plans to release anything faster than 3.2GHz for the remainder of the year. This means that 3.2 GHZ, FSB 1600 Yorkfield might be the fastest Intel for almost three quarters."

But this is not definitive: "Intel is known for changing its roadmap on a monthly basis, and if AMD gives them something to worry about we are sure that Intel has enough space for a 3.4GHz part."

So, in the end we are still left guessing. AMD may or may not be able to keep up with SOI versus Intel's bulk silicon. Intel may or may not be stuck at 3.2Ghz even using 45nm. AMD may or may not be able to hit 2.6Ghz in Q4. However, one would imagine that even if AMD can hit 2.6Ghz in December that only 2.8Ghz would be likely in Q1 versus Intel's 3.2Ghz. Nor does this look any better in Q2 if AMD is only reaching 3.0Ghz while Intel manages to squeeze out 3.3 or perhaps even 3.4Ghz. If AMD truly is the victim of an unmanageable design process then they surely realized this by Q2 06. However, even assuming that AMD rushed to make changes I wouldn't expect any benefits any sooner than 45nm. The fact that AMD was able to push 90nm to 3.2Ghz is also inconclusive. The fact that AMD was able to get better speed out of 90nm than Intel was able to get out of 65nm could suggest more skill on AMD's part or it could suggest that AMD had to concentrate on 90nm because of greater difficulty with transistors at 65nm's smaller scale. AMD was delayed at 65nm because of FAB36 while Intel needs a fixed process for distributed FAB processing. Too often we end up with apples to oranges when we try to compare Intel with AMD. Also, we have to wonder why if Intel is doing so well compared to AMD with power draw then why did Supermicro just announce World's Densest Blade Server with Quad-Core AMD Opteron Processors.

To be honest I haven't even been able to determine yet if the K10 design is actually meeting the design parameters. There is a slim possibility that K10's could show up in the November Top 500 Supercomputer list. This would be definitive because HPC code is highly tuned for best performance and there are plenty of K8 results for comparison. Something substantially less than twice as fast per core would indicate a design problem. Time will tell.

218 comments:

1 – 200 of 218   Newer›   Newest»
Scientia from AMDZone said...

Ho Ho

"We know for a fact that current 3GHz 45nm quads take way below 100W under full load"

Actually, it is a fact that QX9650 is rated at 125 watts. You would have to give some proof showing that Intel's rating is wrong.

"Well, they planned to release at 2.6GHz mid-year."

No. The previous roadmap was 2.3Ghz at release and 2.6Ghz in Q2 2008. Obviously with only 2.0Ghz at launch they didn't meet this.

"Well, comparing supercomputers is a trciky business. CPU speed is not that important so comparisons are mostly meaningless in terms of finding out CPU speed."

No, not true at all. The maximum speeds are given and these are not typically limited by interconnect speed. The Cray architectures in particular should give good comparative results.

"I expected at least something on a similar level as on Roborat's blog"

Some of the commentary on roborat's blog came from Intel employees. Since I don't work at Intel it would be difficult for me to match this. I'm sorry if this fact didn't occur to you. However, the point that you also miss is that none of the commentary on roborat's blog came from AMD employees so it is more than a little one-sided. And, I am sorry if your preference is for a one-sided, pro-Intel discussion. The question that has not been answered is whether or not Intel has gained a clear lead due to RDR.

Erlindo said...

As always, great writing Sci.

I do have this little hypothesis going on in my head for along time and I'd like to know what you think about it:

Barcelona should be able to outperform K8 by a great amount (15-30% in Int apps and 40-80% in FP) according to AMD's benchmarks against a simulated quad core K8.

My best guess would be that the original core was performing the way AMD wanted to, BUT they found a big problem in the core (bug?) which made them pull back from the original release which was supposed to be in the march/may timeframe.

When they (AMD) saw this flaw, they didn't had more options than to disable/cripple some of the processors most powerful features. AMD engineers didn't have more time but to release the processor the way it currently is because another delay would be really bad for AMD's weak finances. To me, The September launch of Barcelona was indeed a delayed launched, but also a rushed job from the guys at AMD.
Correct me if I'm wrong

Common sense tells you that you don't spend almost 3 years designing a new uArch for it to perform equally or up to par with the previous generation. For that case, it would've been better for AMD to currently keep shrinking K8 and clock it as far as they can till they could release Bulldozer, but anyhow, I'm more inclined to believe what you've said about AMD solving this flaw with the 45nm process (Shangai any one?)

I know the above mentioned sounds like a nasty soap opera, but this is just my personal thought of the current situation.

Feel free to correct me if I'm wrong.

enumae said...

Scientia
You would have to give some proof showing that Intel's rating is wrong.


Look at the reviews, the power consumption is equal to 65nm Dual-Cores E6750.

Or is Intel wrong and the rating for the E6750 should really be 130W?

Scientia from AMDZone said...

erlindo

I can only give estimates about the change in speed from K8 to K10. Roughly speaking I would expect 15-25% increase in Integer and at least 80% in FP. It wouldn't surprise me if there were a problem with K10 that hasn't been fixed but I agree that AMD would have no reason to mention this until it gets fixed. For example, Apple admitted that its Mac+ was too slow just as soon as it released the Mac II. The only two ways I know of to know for sure would be to either see a fully tuned FP score that was worse than 75% faster per core or to see AMD release a new revision that was suddenly noticeably faster. In fact, if AMD didn't get it fixed until 45nm I doubt anyone would even suspect that K10 had a problem.

enumae

Good point except you can't test the processor's power rating with Cinnebench and even George Ou should know this. They would have needed to run something like FFT or Prime95 to get even 80% loading. You can't make assumptions about power draw based on the TechReport tests.

enumae said...

Scientia
They would have needed to run something like FFT or Prime95 to get even 80% loading.


Ok, here is another review Bit-Tech, and it is showing the power consumption to be just above (5%) the E6850 which is rated at 65W...

Scientia from AMDZone said...

ho ho

"130W"

Odd that you know what the real TDP rating is and then claim that it is less than 100W.

"I was talking about the pulbicized/leaked one from the beginning of the year, before they changed their targets."

Yes, I'm sure that would compare with the earlier Intel roadmaps that showed C2D reaching 3.6Ghz in 2007.

"Well, I'd say it is rather clear that AMD is quite a bit behind."

AMD was 12 months behind in launching 65nm and are now nowhere near where they should be with K10. The strange thing is that you keep deluding yourself in thinking you are saying something that I don't know. All you do is repeat things I've already said and then pretend that you are saying something different. Keep pretending.

"AMD and its CTI launches much sooner and works for a long time before matching its older processes, not to mention beating them."

You seem to be confused. The discussion that you've referred to on roborat's blog suggested that CTI was now slowing AMD down and causing them to fall further behind. I'm not sure what point you think you are making by simply saying what CTI is.

"You've claimed that AMD should be around 3-6 months behind Intel by 32nm "

You never can seem to repeat what I say correctly. Let me make it a bit more clear for you. First of all, this has nothing to do with what I've said; these are statements from AMD itself:

65nm (historical) - 12 months trailing
45nm (expected) - 6 months trailing
32nm (expected) - 0 months trailing

The expected timeframes are based on AMD's stated goal to deliver both 45nm and 32nm in only 18 months instead of 24 months. Maybe you will remember next time.

"Even if they launch at half a year interval it will take AMD at least a additional year to catch up to its 45nm whereas Intel will likely start from where 45nm left off."

Ho Ho, I'm not actually sure what you are talking about here. I guess it is possible that you are talking about AMD's remarks about maybe using high-K in 2009. However, there is no suggestion that high-K would slow down 32nm.

"Most supercomputer stuff works on doubles, Barcelona doesn't seem to help all that much in that regard, at least not when comparing SSE performance."

I've looked over the HPC scores a number of times and they are clearly indicative of maximum performance. This can be clearly seen with K8 and C2D as C2D has roughly double K8's scores. Since there are no K10 scores in the top 500 list yet any conclusions by you are a bit premature.

"But you can quote Fuad without any proof and claim he is right?"

You don't read very carefully. I never said whether Fuad was right or wrong; I said that he was the same source for both saying that AMD wouldn't have 2.6Ghz until Dec or Jan and for saying that Intel wouldn't go over 3.2Ghz in 2008.

"If a 130W TDP CPU takes as much power as a 65W"

based on a Cinnebench test?? You need at least Prime95 to come up with any valid power draw numbers.

"there is a good chance that the 130W one was rated way higher than actually needed, don't you think?"

You honestly think that Intel is under rating its own processors?

Scientia from AMDZone said...

enumae

Thank you; you've brought up a good point with the Bit-Tech review. The problem with the review is that Prime95 is single threaded. To actually measure the processor power draw they would have had to run a copy on each core. There is no indication that they did that. However, I'll readily admit I was wrong if you can show me a test that was conducted properly.

Mo said...

First of all.... it took AMD 3 Years to get to 3.2Ghz on 90nm process.

Intel was at 3ghz in one year (Core 2)

AMD's 65nm is reaching the Year mark and where do we stand on that?

Also, why in the world would I care if AMD is int he top 500HPC?
How does the help me? how does that help the performance of a single Phenom?

I could care less about HPCs and heavy clusters in tight spaces and what not. It doesn't effect me because that's not what I need.

I need a fast single chiped system. AMD has yet to deliver and and it seems like it will fail to do so.

Greg said...

Mo, if you could care less, why are you even commenting here? I mean, isn't the point of all of this to analyze exactly what's happening with these products?

Ho Ho said...

scientia
"Odd that you know what the real TDP rating is and then claim that it is less than 100W."

Wow. Do you actually know the difference between TDP and power usage?


"Yes, I'm sure that would compare with the earlier Intel roadmaps that showed C2D reaching 3.6Ghz in 2007."

Difference is that AMD advertised K10 with 2.6GHz CPUs.

"Ho Ho, I'm not actually sure what you are talking about here."

65nm K8 launched nearly a year ago. Where is it now compared to 90nm?


"You need at least Prime95 to come up with any valid power draw numbers."

How big is the difference, really? More than 10%? More than 20?


"You honestly think that Intel is under rating its own processors?"

I'd say it is a fact.

"However, I'll readily admit I was wrong if you can show me a test that was conducted properly."


Is this one or this better?

Now tell us why do we have to bring you links showing the power usage if you are so sure that it is 130W? Have you actually seen any evidence that Penryn would use as much power as you claim?

enumae said...

Scientia
Prime95 is single threaded


It depends on what version, I found Prime95 25.3 and it lets me select the number of threads.

For clarification, what is the point to test power consumption and only use one thread?

abinstein said...

scientia -

Sorry but I fail to get a conclusion out of your article. So do you think Intel's process tech is leagues ahead of AMD, or do you not? It seems to me that you did a lengthy discussion without answering what makes up the difference between performance of the two companies:

1) budget in process development?
2) process technology advancement?
3) choice of bulk vs. SOI?
4) circuit design methodology?
5) microarchitecture bugs?
6) fab management?

In my opinion AMD is clearly behind in both 1), 2) and possibly 5) above. I don't think RDR is of anything special to Intel because I know for a fact that similar "rules" exist elsewhere. One of the early arguments against SOI is that it's harder/more restricted to design circuits for it; what we see from RDR is that as dimension shrinks even bulk Si is having the same problem.


Ho Ho & enumae -

TDP is potentially the max power usage under the worst permissible cooling environment. CMOS circuits take less power under lower temperature; most reviews use very good if not ideal cooling, which reduces power consumption much under TDP. The same is not true for general application.

Besides, as scientia said, most benchmarks do not stress all four cores 100%.

Giant said...



It depends on what version, I found Prime95 25.3 and it lets me select the number of threads.

For clarification, what is the point to test power consumption and only use one thread?


The ideal scenario is to open two or four instances of Prime95 depending on the number of cores. Then open a program to monitor the temperature like Everest or CoreTemp. Set each instance of Prime95 to a specific core and then run them all. This will push all the cores to the maximum in terms of heat and power use. It's also very good for testing the reliability of a system after overclocking.

enumae said...

Abinstein
...most reviews use very good if not ideal cooling, which reduces power consumption much under TDP...


I'm sorry but you need to explain this because it almost sounds like you are saying that using a better cooler will allow the processor to use less power while under load.

Giant
The ideal scenario is...


Sorry, it was actually a rhetorical question.

Scientia has/had continued to dismiss the links shown to him that clearly show Intel's QX9650 is using less than 130W under load.

Thanks though :)

abinstein said...

using a better cooler will allow the processor to use less power while under load.

Yes that is what I meant.
Cooling affects power consumption rate. A cooler IC runs further cooler. That is why all IC specify their TDP w.r.t. some (max) temperature.

Ho Ho said...

abinstein
"MOS circuits take less power under lower temperature; most reviews use very good if not ideal cooling, which reduces power consumption much under TDP."

Many sites used default cooling and saw similarly big power consumption drop.

Btw, how big is the difference in power usage when comparing regular inbox with, say, good water cooler? More than 5W for 100W CPU? I highly doubt it.

Peter said...

Enumae said:

Scientia has/had continued to dismiss the links shown to him that clearly show Intel's QX9650 is using less than 130W under load.

More links to Yorkfield power consumption:
http://www.lostcircuits.com/cpu/intel_yorkfield/4.shtml
Equal to E6750 in Cinebench rendering, even better at idle.

http://techreport.com/articles.x/13470/15
Hey Scientia, here are some Prime95 results. And yes, these are obviously from the multithreaded version of Prime95, just look at the wattages of the other CPUs.

Pop Catalin Sever said...

On the process side AMD needs a good .45 nm process. The process needs to fix both leakage and and the maximum achievable speed of current .65 nm process. As AMD has stated before they will use ultra low-K in their .45 nm process which theoretically should improve power consumption, but I'm having a finding meaningful information in this regard.

On the performance side the fastest AMD can regain performance crown, is when SSE5 makes it debut ??? that's in 2009, so this means AMD will have to handle the pressure of Penryn and (maybe) Nehalem in 2008.

Intel is currently giving multiple blows to AMD: it has an better micro architecture that is also capable of hinger clock speeds on .65 and even more on .45 with much improved Performance/Watt, it can produce quadcore chips cheaper because of MCM, it will release it's own direct connect architecture in 2008 practically denying AMD's advantage in HPC. By the time penryn raps to volume Intel can (and will) start a second price war on AMD ...

There is no doubt that there are some key aspects that AMD must focus on and not fail or delay in delivering them: a much better .45 nm process and SSE5. I hope they are fully aware of this and I'm somehow hoping that AMD has the financial and technical power to push those forward for the sake of the company.

abinstein said...

Ho Ho:
"Btw, how big is the difference in power usage when comparing regular inbox with, say, good water cooler? More than 5W for 100W CPU? I highly doubt it."

It has nothing to do with how much you doubt it, which is of no significance in the face of physics.

I don't remember the exact numbers but the difference (of 20C, for example) is much more than 5% power draw.

Mo said...

Greg

Can you analyze the Phenom and get back to me?
Why can't we analyze products that we're ACTUALLY GONNA USE?

Are you gonna use an HPC? Are you gonna build a 4-8Way server for your use?

I'm talking about that HUGE HUGE consumer market.

Let's talk about them.
I'm not liking the initial K10 Phenom results...are you?

Scientia

New versions of Prime95 are multi-threaded.
I'm sure these review sites know how to test for power consumption, it's not their first time.

Look at other reviews like X-Bit, they clearly state they used Prime95 25.3 which can run multiple instances.

Ho Ho said...

It would be much better if you could actually prove what you say.

Also you seem to silently ignore that reviews use the same cooling on most CPUs, that means they all get similar reduction. Higher power using CPUs should benefit even more than Penryn with its little useage.

Ho Ho said...

That last one was meant for Abinstein, I didn't think that someone would post in between

S said...

Has AMD already demonstrated 32nm SRAM wafers? If so it is a big achievement for AMD to be on par with Intel in 32nm. It is commendable.

Ho Ho said...

They still have about 3 months left to demonstrate it if their original schedule of 18 months form 45nm to 32 still holds.

Aguia said...

According to Xbitlabs the QX9650 is consuming 89.8 Watts at load which means it better fits the 130W rating than the 90W for example.
Secondly Intel and AMD always give the maximum TDP for their high end parts (that normally will be overclocked).


I think Intel and AMD should revise their Power ratings.

First: The power rating means nothing unless someone can point me one motherboard in the market that says it’s limited to some of those ratings?

Secondly: The use of some standard application (or applications) that would be used to measure power consuming of the processors and put some rate on the box like in some Household appliances (refrigerators for example) or do the easy way of using the standard lamps ratings.

Maybe with that we could see energy efficient motherboards in the future too.

AndyW35 said...

Thanks for the blog post, some interesting points.

I think you can split architecture bugs from process materials issues. The first can normally be worked around or hidden, but the second could be a real showstopper, or at least a cap.

For instance, I consider the increased cache latency in 65nm K8 to be more likely connected to eeking out the speed of the chip than a "trial" to allow bigger caches in the future, however the 65nm K8's still seem to have a 65nm process speed cap on them.

Similar to this K10 seems to be having a few bugs, but if the speed was well up then these could be hidden, but it seems K10 65nm is also speed capped currently. I'm not convinced they can get 3+GHz out of it to overturn any possible problems with the architecture they have, they are more likely to fix those bugs.

45nm for AMD really does have to deliver the goods, but as the process gets smaller the issues for SOI possibly add up, most particularily in heat disipation on one hand or leakage on another if the insulater layer is kept tight to avoid heat buildup.

So hopefully they can crack this nut and it not be a cap so they are not limited.

I agree with you that Intel are not likely to go anywhere near 4GHz, but their process can be described at least as equal to their previous, something I hope AMD can also achieve. I'm not totally convinced yet they can.

Axel said...

Ho Ho

Wow. Do you actually know the difference between TDP and power usage?

This has already been explained in comments on a previous blog entry. It's really quite simple. Due to low 45-nm output over the next couple months, Intel will be binning the best quad-core Penryns only into the server market. These are called Harpertowns and the 3.0 GHz Harpertowns will have a TDP of 80 watts. Since Intel do not expect to yield enough 3.0 GHz 80 watt 45-nm quad-cores to supply the combined demand for server and desktop, they are expanding the bin split for the desktop (where TDP is less important than server) and allowing 3.0 GHz Yorkfield to have a much higher TDP. As 45-nm output and yields increase through 2008 and new steppings are introduced, Intel will inevitably bring the desktop offerings into roughly the same thermal envelopes as the server products.

Axel said...

To continue, the low power consumption shown in the current QX9650 reviews mean that these were probably cherry picked samples and were probably sourced from the top Harpertown bin (80 watt TDP @ 3.0 GHz). The actual shipping QX9650 parts will likely be sourced from the lower Yorkfield bins and will generally have higher power consumption consistent with their higher TDPs.

enumae said...

Axel
...the low power consumption shown in the current QX9650 reviews mean that these were probably cherry picked samples and were probably sourced from the top Harpertown bin (80 watt TDP @ 3.0 GHz). The actual shipping QX9650 parts will likely be sourced from the lower Yorkfield bins and will generally have higher power consumption consistent with their higher TDPs.


I think you nailed it!!!

Scientia from AMDZone said...

ho ho

Thank you for mentioning the Toms Hardware article. It seems to have what we need. Specifically:

we stressed all four cores using Prime95

So, we have a pretty good initial estimate of TDP. The second step though is to convert these numbers into actual TDP which is higher because Prime95 is not maximum power draw. Based on some of the other information I would give it a power rating of 70%. In other words, part of the reduced power draw is because Penryn is not stressed as much by Prime95 as C2D. This would put QX9650 at:

3.00Ghz - 104 watts
3.33Ghz - 113 watts
3.66Ghz - 134 watts

If you make the mistake of extrapolating the power draw forward based on the other architectures then you end up with an amazingly low power draw that easily fits in 89 watts. However, Prime95 is only as useful as it is able to stress the architecture. If the architecture is efficient enough to do Prime95 with less work (which Penryn seems to be) then Prime95 is not as useful for estimating TDP.

Even so, this does suggest (assuming my estimates are close) that Penryn has enough headroom to hit 3.33Ghz with no problem.

Mo said...

Scientia,

What do you advise should be used to stress the computers to maximum TDP?

Scientia from AMDZone said...

No, apparently I was wrong. I just checked the updated thermal guide (which is a very good reference) and Prime95 is the reccomended program.

So, I would now have to say that the testing was good and that Penryn does show a distinct drop in power consumption versus C2D.

enumae said...

Scientia
So, I would now have to say that the testing was good and that Penryn does show a distinct drop in power consumption versus C2D.


So do you have a belief that Intel could hit 3.8 - 4.0GHz with Nehalem considering the review at Toms Hardware?

Pop Catalin Sever said...

"So do you have a belief that Intel could hit 3.8 - 4.0GHz with Nehalem considering the review at Toms Hardware?"

Well 3.8 was an expected speed for Core 2 on .65 nm too, but that never happened, Intel might push the thermal envelope if pressured like it did with P4 that was throttling at max load, no doubt about that.

The only good answer why it is not doing that is because Intel likes to sell lower bins as extremme CPUs else they would have released higher frequencies even for Core 2.

Hornet331 said...

http://www.theregister.co.uk/2007/11/06/ibm_opteron_x3455/

any comments on that?

it seems that ibm has serious issus to get enough chips... :(

Axel said...

In answer to this blog entry title: Yes, AMD's process tech is well behind Intel's.

Greg said...

Mo, it seems to me the point of commenting on this blog is not to inanely state the performance numbers generated by these products but to figure out what those numbers actually mean.

Scientia's comment addresses the important question of why did AMD's architectural improvements yield so little benefit so far? Is it a system/coding issue or a processor issue? If it's a system/coding issue, the HPC numbers would show a significant performance increase over k8, or at least a performance increase that's closer to what we'd expect the architectures improvements to yield.

Axel makes a good point (on the previous article's comments) that there is some possibility that there's an architectural bottleneck inherent in the basis of k8s design, but I really don't see where it could be (though I admit I'm no expert).

Hoho, actually, just thinking about the basics of the problem, using the same cooler would still cause inconsistencies in the max power draw of a chip, being that the power draw would not increase linearly with temperature, and this would also be exacerbated by the materials used in microprocessors (since we're not looking at copper, which is more ideal).

However, I think all this means is that this is just another problem the product has to overcome by design, so consistency in cooler ability is paramount to unbiased testing. I'd think the best choice would be a decent multi-socket cooler that didn't cool too exceptionally well, but also didn't suck too much, since this would both represent what most people would buy and avoid moving to either cooling extreme.

Greg said...

As a correction (or clarification) to my last comment, I meant that using the same cooler on every processor would not necessarily mean that all the processors maximum load power usage would actually be reached, not that the same processors would be inconsistent with the same cooler.

AndyW35 said...

Digitimes are reporting that 2.2 and 2.3GHz at launch, 2.4GHz in December and 2.6GHz in 2008:-

http://www.digitimes.com/mobos/a20071107PD220.html

I am not sure what time frame, if any, 3GHz Phenoms on 65nm will make an appearance.

AMD will have to make hay in the server side only it seems, the desktop parts being midrange midpriced parts not able to compete with the top performing Intel 45nm cpu's. Which is a shame for high end consumers but for midrange consumers does give more options. I can see the $200 battle-ground being highly fought over.

I still see little need for quadcore on the desktop, if I was buying a new cpu I'd look at higher performing dual cores. The Intel E8400 is suggested to be at $183 which is very good value for a 3GHz 65W part.

Pop Catalin Sever said...

Top Mistakes AMD made under Hector Ruiz:

- Failing to update key parts of the K8 microarchitecture in a reasonable time frame after it's introduction (2 years). I'm taking about integer IPC and updating SSE to do 128bit operations per cycle.
- Failing to deliver an MCM quadcore chip prior to Barcelona.
- ATI acquisition instead of partnership that could provide similar benefits adding debt to company and not using the financial resources in where they would have made a difference R&D and superior fabrication process.
- Failure to switch to High-K metal gates for .45 nm (??)
- Failure to predict and react to competition moves.

Rumors about H.R. resignation have started to surface on the web and hardly anyone can say that they can be simply discarded.

AMD made some good decisions in this time frame but not enough to to counter the mistakes they made.

I would surely like to hear your opinion on why AMD sits where it sits today Scientia.

Mo said...

AMD would have been in better position right now if they had not decided to acquire ATI.

At the same time, No one can predict the future but, that's why analysts for the corp. are payed so freaking much.

They should not have under-estimated the Core 2 arch. That's exactly what AMD did. They figured they could buy out ATI, and still be fine with K8 (they were also hoping that K10 would be on time and perform well).

Everything has basically fallen apart.

intel's price structure has KILLED AMD profits. That AMD did not predict. AMD would never have guess Intel would price their chips that low. To the point where it puts the K8 at Garage Sale prices.

It's just a bunch of bad decisions on top of bad luck.

If your competition is realeasing a product, and they constantly say that it will be that much faster and better. Why in the world would you go empty your pockets?

Hector's gotta go. K8 was practically here and pretty much sold itself, Hector didn't do crap.
Nothing really happened between then and now.... the Only major development and launch under Hector has pretty much been a failure in my eyes.

Pop Catalin Sever said...

"It's just a bunch of bad decisions on top of bad luck"

Luck has got nothing to do with it. It's a business where you need to make the right decisions to win not to roll the big dice.

Mo said...

Well in that case...........

It's just a buncha bad decisions

Giant said...



Hector's gotta go. K8 was practically here and pretty much sold itself, Hector didn't do crap.


Agreed. The first step to AMD's recovery is to ditch Ruiz and promote Dirk Meyer or ask Jerry Sanders to return. His 'market share at all costs' strategy (aka as 'scorched earth!) is killing AMD.

There was that rumor posted on The Register that Ruiz may well be going next week.

http://www.theregister.co.uk/2007/11/08/amd_cancels_gig/

One nasty rumor floating around suggested that AMD canceled the event because CEO Hector Ruiz plans to announce his resignation next week.

Dirk Meyer was appointed to the AMD board of directors last week as well. That's got me curious.

Greg said...

Ruiz is obviously out, but probably more for PR and appeasing the stock market.

I still can't see what AMD should have done differently. If they'd kept doing what they were doing before Core, and kept focusing on the channel, they'd be worse off than they are now in terms of marketshare, and really couldn't be any better off in terms of profitability (drastically smaller marketshare x slightly better asps = the same if not worse revenue than current).

K8 wasn't changed because that was simply their normal development cycle. Like scientia said, they've only got a couple design teams going. I'm guessing one of them probably produced an updated version of k8 for dual core (k9?) while the other one just worked on dual core k8. The team that only did dual core k8 came out with a better product, so that's what AMD used.

Just because their product didn't change earlier doesn't mean they haven't been trying to update the k8 architecture.

Also, a market that is as heavily dependent on R&D as this is always going to have a very strong reliance on a bit of luck. Intel having a horribly leaky and hot 90nm process: bad luck. AMD having a 90nm process that's, even today, very competent: good luck. AMD having, reportedly, a weak 65nm process: bad luck. Intel having an amazing 45nm process: good luck.

It's not like they choose which process will be good and which will be bad. They simply devote a given amount of funding to researching and developing various aspects of the process shrink, and sometimes a group runs into more snags than usual and can't completely eliminate all the problems they have before the process goes into production.

I don't even know why I have to waste everyone's time going over this. It's really some of the very basic dynamics associated with research.

Khorgano said...

Intel having a horribly leaky and hot 90nm process: bad luck.

Please don't spread FUD on things you don't fully understand. I can see why most people would think that their 90nm process was horribly leaky and ineffecient, but the Prescott issue was an ARCHITECTURE problem, not a process problem. It is obvious to see that the Pentium-M's did not suffer the same fate. Likewise, when Intel moved to 65nm, the Pentium D's were still very power hungry and ineffecient while the later C2D's were much better.

If you actually compare power usage from the process stand point, both Intel's and AMD's were virtually equal in power efficiency. If everyone would like a better explanation we can certainly go into more detail.

Either way though, you're point was still on target in that the Netburst arch was a serious misstep and miscalculation on Intel's part as far as its practical scalability is concerned.

Pop Catalin Sever said...

"Either way though, you're point was still on target in that the Netburst arch was a serious misstep and miscalculation on Intel's part as far as its practical scalability is concerned."

Yes it was an misstep not "bad luck" they knew they were going to encounter leakage issues long before they introduced P4 but they still did it because of the crazy MHZ race they were engaged in. That's a clear example of bad choice not bad luck.

Greg said...

Khorgano, I agree. I didn't really re-read my post that time. Thanks for catching that.

Pop, no... When you assume the people running either company are "stupid" as you insinuate, you're belittling the argument and trying to simplify what happened to a point that's just inaccurate. Essentially, FUD.

Pop Catalin Sever said...

"Pop, no... When you assume the people running either company are "stupid" as you insinuate"

I never insinuated that they are stupid...don't know where you pulled that from...

Making bad decisions doesn't make you stupid but rather inadequate for making that kind of decisions.

Please don't put words in my mouth.

Mo said...

whats up with AMDZone?

Can't sign up or do anything there.

someone might wanna leave Chris a message because the "Register" Feature is not working.

Scientia from AMDZone said...

enumae

I would say that Intel will be able to do 3.33Ghz and of course 3.4Ghz isn't much beyond that.

Who knows? With another stepping similar to G0 I would say that 3.6Ghz would be possible. However, G0 took more than a year so that would be after Nehalem's release. By Q4 08 I would say that 3.5Ghz would be about the upper limit but I don't see it as likely.

Pop Catalin Sever said...

"would say that Intel will be able to do 3.33Ghz and of course 3.4Ghz isn't much beyond that."

So you expect Intel to barely raise speed by 10-15% from 3GHz to 3.3GHz 3.4 GHz an not be able to go higher in spite of all tests that show lower power consumption, lower temperature and overclockability to 4.0 GHz on "air" and also a superior transistor technology.

But on the other hand you expect AMD to go from 1.9 GHz to 2.8 GHz (3.0 GHz) in 2 quarters ? that's a 50% clock increase for AMD which you expect but not more than 15% for Intel ...

Giant said...

At XtremeSystems people are hitting 4Ghz with the stock Intel fan. 4.6Ghz seems to be the limit on most CPUs

lex said...

What limits overclocking?

There are many reasons things may crap out and limit overclocking.

1) A product may have lots of design headroom but not get released at higher clock rates. You may ask why that is. As you raise frequency you burn more power. With more power comes a lot more heat. With more heat you get more less reliability. Oxide and metal lifetime get shorter. Thus you can have a product that can run fast but if you can't keep it cool expect it to burn out sooner. Or you need exotic cooling solutions.

2) A process can simply not have enough design margin. A correctly funcational CPU requires logic synchronization across the whole chip. All these logic functions take some finite time to do, then some setup and hold times too. As you run the clock faster and faster at some point there will be some signal that will just be a little skewed too much that it arrive early, too late, or doesn't stay in the right state long enough.

I'm sure both AMD and INTEL have certain frequency targets and specs for all these logic gates. You design in as much margin as you can. You try to simulate all the interactions; voltage droops, cross talk, clock skews. But when you have 10mm and longer trace lines and power grids supporting millions of gates switching you need really good models and extraction tools to insure your design has margin.

In the end you still rely on lots of silicon debug to validate and fix both logic and design margin bugs. At somepoint you are slave to your orginal design frequency. If the original Merom was designe only to hit say 3.6 GHZ for example. I'd expect that on this final steppting INTEL has fixed some missed margins, identified some slow things, but physicall you can't overclock the thing to 6 GHZ no matter what process you put it on. The chip and logic was never design to absorb the margins that 6GHZ would require for setup, hold and delays. So arguing somehow about a chips overclocking to access technology is very naive. It does give you a clue to how much headroom the design could have on a shrink. But again you can't just crank up the frequency as you run into the power and reliability wall.

No difference then pushing your Honda to run day in day out at 100 MPH. Also no amount of tuning will enable that Honda to run at 130MPH, but your porsche designed bottoms up runs there just fine.

Says nothing about honda not being a good car.

sharikouisallwaysright said...

AMD Black Edition at 4 GHz:
http://www.hardforum.com/showthread.php?t=1239834

My conclusion is that Intel is finished once and for all.

;)

Mo said...

AMD Black Edition at 4 GHz:
http://www.hardforum.com/showthread.php?t=1239834

My conclusion is that Intel is finished once and for all.


You are Joking right LOL. because that's Hilarious.

The guy did a suicide shot at 4.0ghz, on Dry ice at -12C with 1.69V

LOL LOL and from that Intel is finished which can do 4ghz on stock air hahaha

Ho Ho said...

scientia
"I would say that Intel will be able to do 3.33Ghz and of course 3.4Ghz isn't much beyond that."

You do know that 45nm 3GHz quadcore Xeons will be sold with 80W TDP, do you? From 80W at 3GHz it shouldn't bee too hard to get way over 3.4GHz with 130W.

Also note that all the quads from 2 to 3 GHz are rated at 80W but you can bet that lower speed ones are actually consuming much less than that.

Aguia said...

ho ho,

You do know that 45nm 3GHz quadcore Xeons will be sold with 80W TDP, do you? From 80W at 3GHz it shouldn't bee too hard to get way over 3.4GHz with 130W.

But is that at standard voltage ho ho?
And how many can they bin?
I’m sure it will hit 4.0Ghz but not in the current revision.


giant,
At XtremeSystems people are hitting 4Ghz with the stock Intel fan. 4.6Ghz seems to be the limit on most CPUs

Giant did you see the voltage they are running at? That’s complete CPU suicide voltages for 45nm CPU’s. Even at 65nm it’s already too much.

Hornet331 said...

i wouldn't call 1.4V much for 4ghz, heck my E6600 B2 needs 1,37V to be stabel for 3,2ghz.

Also overclockers give a crap about lifespan?

How long do they have a cpu 1 year? max. 2 years?

Ho Ho said...

aguia
"But is that at standard voltage ho ho?"

They are selling those so I assume they are running at standard voltages.

"And how many can they bin?"

I have no idea but we should get a rough idea once desktop versions arrive and we can measure their power usage. If it is nearing 130W then not much. If it is <100W then probably most could work at 80W.


"Giant did you see the voltage they are running at?"

I didn't but IIRC Intel has said the CPU survives up to 1.5V just fine. Btw, how short would CPU life get if it gets 20% more voltage than in standard? 5 years instead of 15?

Giant said...

From XtremeSystems:

This is nothing.....

It can do 4374MHz 2k6 with ONLY 1.36VCore....


4.653Ghz with 1.488V. That's nothing for such a highly clocked CPU. My B3 stepping Q6600 needs 1.4V to do 3.2Ghz. I run it at 3Ghz 24/7 though.

lex said...

Should overclockers care?

They had better even if they only use the computer for a year or so. If they choose to overclock with voltage even something so small as a few hundred millivolts! Especially if they don't take extra effort to cool their part.

I'll wager these overclockers aren't your average joe so not only is your CPU running voltage higher its average workload and activity factor on the chip are far higher then the "average" usage baked into "average" lifteime. You add a few 10's of degree in transistor temperature and you can easily lose 10x lifetime. Last I checked most manufactures spec to no more then 10 years on these parts.

Don't come crying when your overclocked AMD part dies in 6 months.. Well at least then you can correct your errors and get a cool running Penrym 45nm that won't have the gateoxide crap out at 125C.


http://www.ece.rutgers.edu/~kpckpc/Papers/NanoSymp2001.doc

Scientia from AMDZone said...

lex

Your post was quite long at 2,500 words. Unfortunately your insult and expletive to fact ratio was too high so I had to delete it.

You make a few good points but most of your post is either inacurate (like your attempts to describe process history) or it shows your misunderstanding of what I said.

I don't think I can reply to all of your mistakes about process here but that could be another article.

Let me see if I can hit some points anyway.

There is no doubt that Intel is ahead with 65nm. Intel released 65nm a full year ahead of AMD (Q4 05 for Intel and Q4 06 for AMD). Secondly, there was no doubt that Intel's 65nm process produced both higher clock speeds and lower power draw (compare smithfield to presler). AMD, however has yet to match even 3.0Ghz with 65nm while its 90nm is reaching 3.2Ghz. AMD has some work to do.

Secondly, with 45nm there is no doubt that Intel's 45nm process exceeds even the G0 stepping of its 65nm. However, your statement that Intel is first to 45nm "by a year or more" is pure fiction. This would only be true if AMD does not release 45nm until late Q4 2008 (which of course has not happened yet). The only statement we can make so far is that Intel is roughly on track with 45nm arriving two years after 65nm and that Intel is ahead of AMD. Following AMD's statements, the lead is at least six months.

I'm sorry the mention of FinFETs confused you so I'll see if I can simplify. A few years ago (2002) when IBM and AMD demonstrated FinFETs these were suggested to be the future of digital architecture:
"The superior leakage control characteristics make FinFET transistors an attractive candidate for future nano-scale CMOS generations, which are expected to be in manufacturing within the next decade."

Intel (if you'll recall) demonstrated its own Tri-gate transistor:
"Compared to today's 65nm transistors, integrated tri-gate transistors can offer a 45 percent increase in drive current (switching speed) or 50 times reduction in off current, and a 35 percent reduction in transistor switching power."

The problem is that at this point there has been no official statement by either AMD or Intel about when (if ever) these will actually be used. So, these non-planar devices could be irrelevant, just the next step in gate design, or a big technological leap.

And, finally, I'm not sure how you got confused about RDR. For example I never said or suggested that RDR had put Intel behind in anyway. In fact you have what I said backwards since I was suggesting that Intel would have had to have known that there was a design problem with Prescott (if they weren't aware earlier).

Greg said...

I was insulted by the fact that in all of those 2.500 words lex hand't said anything he hasn't already said except for maybe one or two paragraphs worth of material.

If lex actually respected our intellect he would have bothered being concise. I realize this is also something I have a problem with, but at least I don't repeat myself without apologizing.

Also, mo, if lex bothered taking a really long time to write that, I feel more sorry for him before it was deleted than after.

Azmount Aryl said...

mo said...
The point is some of you have no respect of other's opinions and points of views.



And what if lex is working for Intel in their viral marketing team? Shouldn't then the blog owner reserve the right to limit his activity?

Aguia said...

I didn’t read lex post because it was too long and from one or two paragraphs I read pass he seemed to say the usual things.

However my self have been victim from the Sci del a few times and I didn’t really like it.

It seams Scientia must say what rules he bases for DELETION. Offending? Name calling? Misleading information? Someone opinion with wrong information?

Some times we take much time to make the post and to arrange the information inside the post, we should have at least the right to retype some part of the info in case of something is wrong.

Aguia said...

To ho ho and the others including Scientia, that keep saying that Intel could do some amazing 90W TDP 3.2Ghz CPU at 45nm.
Not really...

http://www.techreport.com/discussions.x/13577

I wonder if the way Toms measure power consuming is the correct one. I think the Xbitlabs measures are more accurate. However to bad there are no others sites that measure CPU power consuming to have more results than just two web sites.

Ho Ho said...

"I think the Xbitlabs measures are more accurate"

Too bad they didn't measure OC power usage. Tom did and even at much higher than 3.2GHz it was quite little more than at 3GHz.

Aguia said...

ho ho,

Too bad they didn't measure OC power usage. Tom did and even at much higher than 3.2GHz it was quite little more than at 3GHz.

Yes ho ho that would be cool.

Another interesting test (for me at least) would be at the same clock speed increase and decrease the CPU voltage to see the impact on the power consuming and/or heat.

Giant said...

AMD in talks with TSMC to sell FAB30/38.

http://www.eetimes.com/news/semi/showArticle.jhtml?articleID=202805354

It's only a rumor, but I did predict that AMD might sell off FAB30 and make do with FAB36. They can outsource CPU production to TSMC if need be.

AMD_FAN said...

Maybe they got that wrong.

AMD will indeed sell the 200mm tools and with that money they can sudsidize the 300mm tools for Fab-38 (I wonder if they did already).

The other question is when will AMD start building the NY Fab, and if they choose to do so, will they use 400mm tools on it?

Giant said...



The other question is when will AMD start building the NY Fab, and if they choose to do so, will they use 400mm tools on it?


AMD does not need more than two fabs. They're in a bit of financial trouble now too; cash isn't exactly in abundance at AMD now. If AMD did build the NY fab it would be using 300mm wafers. 450mm wafers are set to be used until some time next decade.

Axel said...

Giant

AMD in talks with TSMC to sell FAB30/38.

This would be the first logical step in Asset Light. I bet that the rumor has more than a hint of truth to it.

Ho Ho said...

"AMD will indeed sell the 200mm tools"

... to Russia, not to TSMC. Also I think that transaction has already been made.

Aguia said...

Well with so little production and with so little market share AMD only needs one factory.

I think the only mistake AMD made was not to buy ATI but to build another factory when one was more than enough with the chartered partnering.

The two factory scenario is only good if ATI could do at AMD what it does at TSMC/UMC/...
Unless AMD could build Cell processors or other IBM CPUs, but the problem is, is IBM not having enough capacity for all it builds?

Aguia said...

AMDZone is pretty much DEAD now.

Maybe they were just "idling" for the complete merge of AMD with ATI that was rumored to take as much as one year to complete:

Nice box

Erlindo said...

Axel wrote:This would be the first logical step in Asset Light. I bet that the rumor has more than a hint of truth to it.

...And
aguiaWell with so little production and with so little market share AMD only needs one factory.


Well, if the above mentioned is true, why is this happening??

AMD Penang expansion targets Barcelona deliveries

Quote
AMD IS EXPANDING its Malaysian assembly plant over at Penang.

Hector's Asset Light plan was obviously some sort of smoke screen, because after all that chat about spinning off fabs and assembly plants, DAAMIT is now expanding its manufacturing facilities.

AMD's Penang facility has 500, 000 sq feet of space, and the plan is to expand the facility by an additional 12 per cent.

Ho Ho said...

Perhaps they just finish it before selling/renting it out

Aguia said...

Erlindo,
Well, if the above mentioned is true, why is this happening??

Because the Barcelona die is too big ;)

Just kidding :)

Axel said...

Erlindo

Well, if the above mentioned is true, why is this happening??

AMD Penang expansion targets Barcelona deliveries


The Penang facility is an older tech fab that used to make flash memory for Spansion, and also tested and assembled AMD processors. To my knowledge, it does not have 65-nm SOI capability. If I had to guess, it looks like AMD will be using it for chipset manufacturing.

Real said...

To my knowledge, it does not have 65-nm SOI capability. If I had to guess, it looks like AMD will be using it for chipset manufacturing.

Ah no, they will be using it for test and assembly. It is only a test assembly facility and not a fab by any means - it is a low tech packaging house only.

Scientia from AMDZone said...

mo

Thank you for offering your opinon about lex's post. I should tell you though that your opinion of what is passable here is not very important to me.

aguia

The rules aren't that difficult.

1.) Starting off your post by insulting me (as lex did) basically loses points right away. That isn't a very bright way to start a post. Following this up with additional insults (as lex did) will almost certainly get your post deleted. Finishing by insulting me (as lex did) after starting by insulting me is foolhardy at best. You are basically begging to have your post deleted.

2.) Throwing in gratuitous expletives (as lex did) is enough all by itself to get your post deleted.

3.) Trolling will get your post deleted. These are the typical "brand x sux; brand y rulz" type posts.

4.) The longer your post is the more informative it should be. It can be filled with unfavorable benchmark scores, unfavorable quotes, and unfavorable opinions. It can even be filled with inane comparisons, historical errors, and other factual errors (as lex's was). Your post can even contain a lot of repetition (as lex's did) although I would prefer that this be minimized. However, a post of over 2,000 words (as lex's was) should be absent not only of insults and expletives but weaker trolling language as well.

It isn't that difficult to see lex's milder trolling. Here is a list of words associated with each:

Intel - lead, better, performance, advantage, first, better, leadership, lead, better, leadership, leader, ahead, superior, superior, competent, superior, untouchable leadership, successful, leagues ahead, earlier, earlier, advantage, ahead, maximize, optimize, superior, superior, first, performance, leadership, further ahead, advantages, focused, power/performance, healthy, well.

AMD - lagged, weakness, handicapped, inferior, late, behind, hampered, noise, propaganda, late, propaganda, fanbois, behind, behind, behind, spinning, behind, boring, late, inferior, problem, penalty, late, behind, ills, lagging, disaster, slow, handicapped, inferior, loses, mistake, dumb, grandiose, wasted, failed, finished, BK, loses, bad, broken.

Basically, lex's post was deleted for not one, but multiple reasons.

Scientia from AMDZone said...

giant and axel

There has been no indication (other than rumor) that AMD plans to sell FAB38. On the other hand, an AMD engineer confirmed that enough tooling will be installed to qualify FAB38 at low level production on 45nm.

This supports what I've said.

amd_fan

I have to admit that I have doubts about AMD's ability to start another FAB project in NY. I can't see this getting off the ground unless they get the NY money up front and can start construction without needing any of their own money until 2009. Remember that AMD had to get a grant from Germany just to finish out the FAB36 toolset.

AMDZone

AMDZone is certainly having problems. I've had trouble logging in twice. For example, it shows me as guest but then when I try to log in it tells me I'm already logged in. At any rate, this prevents me from posting. I assume Chris Tom is working on it.

I don't think AMDZone is as dead as it looks though. There seems to be a very high number of lurkers compared to posters. For example, one day I posted a new thread and it had 126 views with no replies before the end of the day.

Scientia from AMDZone said...

abinstein

I don't think there is any evidence that AMD will either sell or rent space in FAB 38 to TSMC. In fact, I'm not even sure that AMD could afford to devote any capacity to doing bulk test/prep runs for TSMC produced items. However, I suppose it would be theoretically possible for AMD to do design proofing for ATI. I don't know of anything that would prevent running bulk wafers through the 45nm tooling. I don't know if this would save any time or if they already do these tests at TSMC.

abinstein said...

"Another interesting test (for me at least) would be at the same clock speed increase and decrease the CPU voltage to see the impact on the power consuming and/or heat."

I don't know what's to fuzz about on this tdp thing. If those chips are cherry-picks, then no measurement on them whatsoever gives you accurate information about the actual releases.

Plus, if two power rate measurements are done under different cores temperatures, then they are not comparable to each other. Temperature affects Ioff (search "subthreashold swing"), and active power usage, too (search "temperature decreases" and see Figure 8, where a slower slope means the transistors will remain in active region longer and thus consumes more dynamic power).

Any IC TDP spec specifies temperature under which it is measured. It is the only scientifically meaningful way of measurement. (Not to mention using cherry-picked chips as sample pool.)

abinstein said...

scientia...

Interesting it's as if you knew I was coming to visit your blog (after a whole week of missing it) and make that reply to something that I didn't even say in this comment area. Anyways.

I also don't think AMD's going to sell FAB38 to TSMC, probably not "rent" it, either. What I said was that it makes sense for AMD & TSMC to jointly operate the fab. AMD got things that TSMC would very much like - SOI experience, fab management, and a team of highly-skilled workers working on a well-built fab. OTOH, TSMC got cash.

IMO, it's not even AMD doing test runs for TSMC produced items, but the other way around. For TSMC fab30->fab38 is so ideal that you just put in 45nm tooling and boom it can ramp to full production in 6-12 months. How can TSMC do anything better than this elsewhere? For AMD they use their intellectual property in exchange of physical resources.

abinstein said...

The large version of Figure 8 of Taur's paper.

Scientia from AMDZone said...

abinstein

Interesting it's as if you knew I was coming to visit your blog after a whole week of missing it) and make that reply to something that I didn't even say in this comment area.

Yes, I know; you said it on AMDZone. However, lately I've found it somewhere between very difficult and impossible to login and post at AMDZone. So, I guess here will do.

"it makes sense for AMD & TSMC to jointly operate the fab."

I'm not sure why that would make any sense.

" AMD got things that TSMC would very much like - SOI experience,"

I doubt TSMC does much SOI work. And, whatever it does do it would have to create its own tech. AMD can't share IBM's process tech.

"fab management, and a team of highly-skilled workers working on a well-built fab."

TSMC currently operates two 200mm FABs and two 300mm FABs. TSMC does the equivalent (in 300mm wafers) of 300K wspm compared to less than 20K for AMD.

OTOH, TSMC got cash.

OTOH, TSMC just got through buying older 200mm tooling from ATMEL; perhaps that is their only interest in FAB 30, purchasing old 200mm tooling.

"IMO, it's not even AMD doing test runs for TSMC produced items, but the other way around. For TSMC fab30->fab38 is so ideal that you just put in 45nm tooling and boom it can ramp to full production in 6-12 months. How can TSMC do anything better than this elsewhere?"

TSMC is till ramping its own FABs.

FABTech, April 2007:

Both of TSMC's operational 300mm fabs will be ramped more significantly in 2007 compared to 2006. Fab 12, TSMC's first 300mm fab, has continued to expand through fab extensions and is now projected to reach 73,000wspm by the end of the year - a 22 percent increase in wafer starts for the year.

Fab 14, TSMC's newer 300mm facility, is expected to ramp to 58,677wspm by the end of 2007, a 40 percent increase in capacity from the beginning of the year when wafer starts averaged 34,333 per month.


In other words, in 2007, TSMC will ramp its 300mm capacity by about twice the total capacity of FAB 38. TSMC also stated back in 2005 that they would build a third 300mm FAB on the same location. I assume it will be ready to ramp when FAB 14 tops out.

Scientia from AMDZone said...

aguia

"Well with so little production and with so little market share AMD only needs one factory."

This is not exactly accurate. AMD only needs one FAB today and AMD only needs one FAB as long as FAB36 can ramp. However, unless AMD's demand does not grow then AMD will use up the remaining capacity of FAB 36 before 2008 is over. AMD definitely needs to have FAB 38 ready.

"I think the only mistake AMD made was not to buy ATI but to build another factory when one was more than enough with the chartered partnering."

I think you are badly underestimating AMD's volume. AMD currently has about 23% of the total market. For FAB 36 to be enough AMD would need to allow its share to decrease over time.

"The two factory scenario is only good if ATI could do at AMD what it does at TSMC/UMC/..."

I assume you are suggesting that AMD makes ATI chips in-house. This is possible but AMD won't buy tooling to do this since the margins on graphic chips are lower.

"Unless AMD could build Cell processors or other IBM CPUs, but the problem is, is IBM not having enough capacity for all it builds?"

This again would make no sense. IBM is obligated to pass along any surpluss demand to Chartered. And, I can't see AMD waiting for Chartered to need more capacity.

Scientia from AMDZone said...

lex

Since you have so much trouble understanding basic ideas let me simplify for you.

1.) Intel's 3.0Ghz processors are currently faster overall than AMD's. The only area where AMD was competitive was 4-way but this has been degraded with Intel's quad FSB chipset for Tigerton.

2.) Intel should be releasing faster 3.16Ghz chips soon. AMD is probably going to be limited to 2.6Ghz this year.

3.) Intel's 45nm process draws less power than their 65nm process. It appears that this will give Intel some headroom to increase the clocks more than 3.16Ghz.

4.) AMD has lost roughly half a billion dollars each quarter since Q4 06. This is about $2 Billion so far. Obviously, AMD cannot continue to do this indefinitely.

5.) AMD's lack of chips faster than 2.6Ghz does not suggest high yields or good bin splits. The offiical comment of "tens of thousands" of Barcelonas in Q3 is also only a tiny volume compared to AMD's total volume of more than 13 Million chips for the quarter. If it were 50K Barcelonas, that would be less than 0.5%.

6.) AMD is outfitting and ramping FAB36 slower than orginally planned.

7.) AMD has not demonstrated operational 45nm chips yet so it is impossible to judge whether they are actually on schedule.

8.) In comparison of Intel's 45nm with AMD's K10 they seem to be similarly late in terms of expected release. For example, late Q3 and Q4 instead of early Q3 and Q4. However, Intel is delivering clock speeds above its current offerings while K10 is considerably below. This would therefore be better execution on Intel's part.

Now, it is not necessary to keep repeating these things over and over while pretending that you are saying something new. The things I've just stated should be known to everyone here.

I am only aware of two current bright spots for AMD:

1.) Intel is likely to hit maximum capacity in Q4 and Q1. This would tend to prevent Intel from putting additional pressure on AMD. Intel however should be caught up by Q2 2008 when volume seasonally drops off. Intel could then increase pressure again.

2.) AMD's current volume share is slightly above the average for 2006 and just under AMD's all time high.

However, maintaining high volume while losing money is not a clearly defensible business strategy. It also seems that things will be tougher in Q2 and when Nehalem is released in Q4. It also isn't clear how much Shanghai will help AMD.

If you have something to say beyond these things and can say them without insults then feel free.

Ho Ho said...

abinstein
"Plus, if two power rate measurements are done under different cores temperatures, then they are not comparable to each other."

Generally nobody cares at what temperature the CPU runs at. Older Voodoo VPU's ran at well over 100C and nobody worried. What some people do care about is how much power the chips use. When you compare different chips then it is meaningless to try to compare them at the same temperature. If you want to have a meaningful comparison you'll compare them at their normal working state. Otherwise you have no ideas how they act in a real world.

In short you dismissing the TDP numbers because Penryn runs at lower temperature is just a failed attempt to show the reduction in power usage not being as good as it is.


"The large version of Figure 8 of Taur's paper."

A nice 200K difference. It would be interesting to know how big is the difference between, say 45C and 60C. Also I remind you that with high-k leakage power is reduced. Also not all parts of chip run as hot as others.

Aguia said...

Thank you scientia and abinstein for the replies.
Since I agree with everything you guys said... no comments.

Aguia said...

ho ho,
So you are saying that Intel cherry picks its high-end 80W Xeons even though it is confident it could release 4GHz if needed?

Don’t tell me that you think Intel increases the power consuming on their chips on propose?

The fact that Intel has one 3.2Ghz CPU manufactured at 45nm with 150W TDP is Intel way of creating new market segments, but if was AMD doing it is lameness.

And don’t tell me that Intel could really create a 4.0Ghz quad core CPU, just won't do it because it does not need to. There are applications in my work that would be great to have a 4.0Ghz CPU even if we have to pay 3000$ for it.
Besides the server market never had a ceiling price point.

abinstein said...

Ho Ho -
"In short you dismissing the TDP numbers because Penryn runs at lower temperature "

You have serious comprehensive problems. I am not dismissing the TDP numbers, but the power measurements. It's you who are dismissing the TDP numbers, just because somehow a 120W TDP chip is measured to take just 89W under some condition that you don't even know about.

Are you from the middle ages? It seems to me scientific method is something you have trouble getting at.

enumae said...

Abinstein
I am not dismissing the TDP numbers, but the power measurements.


Take a look.

I am only using reviews that have the same platform for both the QX6850 and the QX9650.

It shows that the QX9650 is using about 25W's less at idle and about 50W's less under load.

If both chips are rated at 130W why is their a difference of about 50W's in power consumption under load while using the same platforms?

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

If you see an error please point it out.

Thanks

sharikouisallwaysright said...

Scientia, as ATI is now AMD, will you write something about the launch of the 3870 and 3850 GPUs?

I am very positive about them and i am thinking they will sell like hot cake, as the 8800gt would do if available or even better.

This is what the market was waiting for!

Aguia said...

Strange… nobody is talking about another perfect product from Ati, the 3800 series cards.

Good products from ATI/AMD total silence.

"Bad" products from ATI/AMD lots of talk/"publicity"/viral campaigns.

enumae said...

Aguia
Strange… nobody is talking about another perfect product from Ati, the 3800 series cards.


Sorry, but the only thing that is perfect is how they priced it, below Nvidia's offerings.

Good products from ATI/AMD total silence.

Most already knew or had a good idea of how ATI would perform when compared to Nvidia's offerings, so their is nothing really to discuss.

"Bad" products from ATI/AMD lots of talk/"publicity"/viral campaigns.

Your "perfect product" statement is an opinion that is with out the support of most of the review sites.

The launch of the 38** series was not very exciting, but what will be interesting are the reviews that come out on Monday (Phenom) and the future reviews using CrossfireX.

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

No takers on my power consumption post... :)

John said...

Remember when Scientia said that AMD was fooling us with its SPEC scores and that those were simulated K8 quads and not Barcelona?

lol

Remember when Scientia said AMD would need ~2.8GHz to match Intel's 3.3GHz?

lol

Aguia said...

Sorry, but the only thing that is perfect is how they priced it, below Nvidia's offerings.

Just the price?
I wonder why features like DX10.1, HDMI and UVD are not important.


Most already knew or had a good idea of how ATI would perform when compared to Nvidia's offerings, so their is nothing really to discuss.

Agree. However it would be more "Most already knew or had a good idea of how ATI would perform because of the 2900XT so there is nothing really to discuss."


The launch of the 38** series was not very exciting, but what will be interesting are the reviews that come out on Monday (Phenom) and the future reviews using CrossfireX.

That must be because of Nvidia paper launched card, otherwise would be the web killer of the year!


Your "perfect product" statement is an opinion that is with out the support of most of the review sites.

Really... by most you mean only Toms right?

Let’s check the best game site on the web:


Demand for the GeForce 8800 GTX wasn't on the decline when Nvidia released the GeForce 8800 GT. In fact, GTX chips were still on allocation, and Nvidia didn't have enough chips to fill customer orders. A company doesn't voluntarily give up hundreds of dollars in profits per unit unless there's a very good reason. Maybe Nvidia figured out that it could make more money by reducing prices and selling more chips. Or maybe AMD spooked Nvidia with rumors of a new line of Radeon GPUs that are half the size of the original Radeon HD 2000 GPUs but just as powerful. Judging by AMD's new ATI Radeon HD 3850 and ATI Radeon HD 3870 graphics processors, we're leaning toward the latter.


As expected, the ATI Radeon HD 3870 performs almost exactly like a Radeon HD 2900 XT--very impressive considering that the HD 3870 will sell for nearly half the price of the HD 2900 XT. The GeForce 8800 GT wins all the single-card tests with the exception of BioShock. The results shouldn't be surprising because AMD is pricing both new Radeon cards below the GeForce 8800 GT. We only had a single ATI Radeon HD 3870 card, but we did have two ATI Radeon HD 3850 cards for CrossFire. The single Radeon HD 3850 thrashed its GeForce 8600 GTS counterpart in the sub-$200 competition. Our dual Radeon HD 3850 cards performed well in 3DMark06, BioShock, and Company of Heroes, but we could not get the pair to work properly in World in Conflict and Crysis.

The Radeon HD 3870 512MB looks like a winner in the sub-$225 price category. It's slower than the GeForce 8800 GT 512MB, but it's more affordable and offers a similar performance-per-dollar ratio. Expect the battle for the $200 video-card market to get even more intense as Nvidia prepares to release a more affordable 256MB GeForce 8800 GT. The Radeon HD 3850 might be slower than both the Radeon HD 3870 and the GeForce 8800 GT, but you can't argue with the $179 price, and it crushes the GeForce 8600 GTS, its closest competition in the benchmarks.

Aguia said...

john,

Thanks for the link!

Amazing the AMD runs at the same speed of the Intel system.
The gap is closed.

Now its "just" ramp up production and clock speeds.

"Conclusion

We have seen how the board performs with a Phenom GP-9600 processor. Although some of you might be a bit disappointed with the results, that is largely due to the processor used. In fact, If you look at the benchmarks of games, the numbers aren't that far off from the Intel platform. The only more noticeable difference is in the multimedia encoding. In fact, the performance levels of the phenom was better than I expected."

enumae said...

Aguia
Just the price?
I wonder why features like DX10.1, HDMI and UVD are not important.


Yes DX10.1 is new, but please read...

1. When asked about the advantages of picking up a DirectX 10 graphics adaptor today, versus waiting for NVIDIA or AMD DirectX 10.1 products, Microsoft's senior global director of Microsoft games on Windows, Kevin Unangst, replied, "DX10.1 is an incremental update that won’t affect any games or gamers in the near future."

Kevin Unangst

2. "From what I understand, even if you had the next generation of 10.1 hardware, it would be too slow to use the features. You would have to wait two more generations in order to get a real benefit from it. Remember, Matrix introduced environmental bump mapping almost 6-7 years ago? Normal Mapping and bump mapping just made it in the four years since then. In fact, Far Cry was the first normal mapped game to ship. When you look at it this way, it’s the same as 10.1. 10.1 will become actual or three years from now. But not now."

Crytek’s Cevat Yerli

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

HDMI is available fromm both Nvidia and ATI/AMD board makers.

ATI UVD and Nvidia PureVideo HD are already available on the 26** and the 86** series.

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

That must be because of Nvidia paper launched card, otherwise would be the web killer of the year!

Paper launched? They are on Newegg, although they were sold out rather quickly, I think about 2 days, they are there now.

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

I would like to clarify this statement, Your "perfect product" statement is an opinion that is with out the support of most of the review sites., When I said this it was in regards to the performance of the 3870 vs the 8800GT and I should have been more specific. I am not looking at the lowend cards, but I do understand that is where the majority of sales are.

lex said...

Some other obvious points missed over and over and over and over and over and over again.

9) INTEL 65nm raw transistor performance at a given leakage is superior to AMD's/IBMs, TSCM and most other companies. Thus if it was possible to run the same design on either process and assuming same design rules the same part would be anywhere from 5-20% faster or lower power on the INTEL process. WE can't gauge the design rule penalty or advantage one 65nm might have over. But I don't believe any design rule or DFM rule could take away a 10-20% advantage that INTEL process delivers.
Thus if and when INTEL every starts doing graphics this advantage will overwhelm both nvidia and AMD-ati

10) INTEL raw process advantage at 45nm given not only their demonstration, but the release of production CPUs increases the fundamental power/performannce advantage by probably another 30% over the AMD 65nm process giving them a whopping > 40-50% advantage. When AMD finally migrates to 45nm the best they can hope for without HighK / MetalG would be to be again behind by 20-30%

I'd say given listed advantages why do you even question why INTEL is leagues ahead?

Something has to give or a radical change is required as AMD can't continue. Here is my list from most likely to least likely outcomes for the next 3 years.

a) AMD continues to limp along falling further and futher behind. Won't lose billions a year but enough that they continue to be more and more marginalized till they are irrelevant, much like what happened to PowerPC, Alpha, and / or Transmeta

b) AMD somehow finds a white knight with 2-4 billion cash to burn for the next 4 years with the sole purpose of taking business away from INTEL. With infusion AMD holds grounds and catches up at 22nm. ( looks like they might have found a arab sucker for a few hundred million, boy will those oil shieks be in for a shock when they see it takes an investment like they just made at the rate of every quarter for AMD to catch up )

c) AMD folds up and changes business direction in the next 6 months

d) INTEL returns to its old poor exeuction days and fails to excute 45nm ramp, 32nm development, slips nehelem, and/or both of their 32nm Tick/Tocks

e) INTEL missed something on 45nm HighK/MetalG collapses

Either way the answer to your question is Yes INTEL's process technology has put them Leagues ahead!

What is there to argue, AMD's business plan is broken as it is predicated on failure of your competition and not anything AMD directly controls.


Oh one last question. I'm total agree with you I have a difficult time understanding basic concepts. ( Only insulting ME now!!!!) Could you clarify What did you mean in your initial comment here? Remember you really have to simplfy it to me. Those complex analogies to how RDR relate to recording is just way to complicted for me.


" the theory being thrown around these days is that Intel saw the error of its ways when it ran into trouble with Prescott. Then it scrambled and made sweeping changes to its R&D department, purportedly mandating Restricted Design Rules so that the Design staff stayed within the limitations of the Process staff. The theory is that this has allowed Intel to be more consistent with design and to leap ahead of AMD which presumably has not instituted RDR. The theory also is that AMD's Continuous Transistor Improvement has changed from a benefit to a drawback. The idea is that rather than continuous changes allowing AMD to advance, these changes only produce chaos as each change spins off unexpected tool interactions that take months to fix."

Design rules last I checked are things like how large or small or space between features. How much something like a Contact has to be surounded by metal. Or something like how much or little density is allowed.

You are saying that it was design rule differences that doomed Prescott and that different design rules on Cedar Mill and Merom was what set the Core2Duo apart?

Mo said...

Aguia Wrote
"Thanks for the link!

Amazing the AMD runs at the same speed of the Intel system.
The gap is closed."


The OCW review is BOGUS.
There is NO SUCH thing as a 230x10 Q6700. Thus the review is invalidated. Not only are they using the Q6700 out of spec but it's been downclocked to 920FSB!...

and 2nd...The system is HEAVILY bottle necked by video cards....what did they use? pair of 2600 ATI's LOL. It is very obvious that all their game benches were bottlenecked by the graphics. Even a K8 would have matched those numbers.

If you look at the 3Dmark, you can see that Intel is on top most likely because of CPU score.

Review is a crap shoot because the details are omitted like what the individual 3dmark scores were.

I have seen ENOUGH Phenom scores...and none look good.

You will be VERY disappointed come Launch day.

The_Wolf_Who_Cried_Boy said...

In regards to proces nodes I found this artical posted at the Aces liferaft interesting as it explained manufacturing technology in a fashion I could actually grasp (mostly).

http://www.spectrum.ieee.org/oct07/5553/1

Conclusions I drew from it are;
AMD staying with silicon oxide as the gate dilelectric at 45nm is going to severely cap clockspeeds from excessive leakage without resorting to finfets or whatever.

The SOI camp is pursuing metal gate technology in a different direction to Intel, and that the path they have chosen is proving more difficult to manufacture as it's something which is "being considered" for second generation 45nm product rather than the intitial release.

I recall reading some time ago that fully depleted SOI would require atomic layer deposition to be manufactured, and at the time this would be a daunting hurdle to overcome. From the above artical it appears Intel is now using this, does this mean fully depleted SOI is now a realistic option over PD-SOI (I recall it offers some advantages but without a quick google don't know exactly what).

Appologies if this post seems somewhat pro-Intel given where I'm posting it.

lex said...

Two more articles to read

http://www.eetimes.com/news/latest/showArticle.jhtml;jsessionid=2YNOWUYKJV1CSQSNDLPCKH0CJUNN2JVN;?articleID=202806020

http://www.chipworks.com/blogs.aspx

The issue with fully depleted SOI is that the transistor characteristcs for 32nm or less node requires thickness control of the silicon layer to be very tight. There are few methods that do this well. ALD is one, the problem is that the layer we're talking about has conflicting needs, very good thickness control, yet also relatively thick.

Using ALD for HighK requires you to do atomic level growth of something like 20-30 atomic layers. That is a far thinner layer then what is required for the silicon conducting layer.

Regardless, its interesting that INTEL caught AMD and IBM with their pants around their legs on this one.

Everyone prior to INTEL's annoucment were talking about the issues with HighK and then INTEL told everyone it had done it. They have a good few years lead time, pretty much like what they did with strain.

AMD may have had better design due to a desperate silicon technology sitatuation.

With 45nm arrival and a good Penrym and Nehalem Tick-Tock intel has put togather superior silicon with huge manufacturing capacity and is now using that to manufacture a good design.

Good design + great silicon + huge manufacturing

is competing against

Good design + poor silicon + small capacity

Oh... its in a business that takes billions of dollars to play...

Anyone want to predict the winner and the loser?

The_Wolf_Who_Cried_Boy said...

Thankyou for a quick reponse Lex, I couldn't get your first link to work though.

As for winners and loosers, I know it's cliched but as countless others point out the winner in a monopoloy market is never going to be the consumer.

I guess my first post boils down to ; what is Shanghai going to bring to the table.

One advantage AMD may have if this is anything to go on;
http://eetimes.com/news/latest/showArticle.jhtml;jsessionid=2DIS5VAKVX25EQSNDLSCKHA?articleID=202806020
is the use of immersion lithography to minimize die area.
To quote;
Perhaps surprisingly, Panasonic achieves tighter metal pitches than Intel. While Intel might be proud of extending dry litho to 45 nm, it cannot match the dimensions from Panasonic's fab, which is running immersion tools now. For example, the UniPhier device displays a minimum pitch of 138 nm up to metal four. Compare that with Penryn's metal-two pitch of 158 nm.
While I assume keeping having a traditional transistor structure will bring limited increases in available clockspeeds, Shaghai would be a much samller and vastly more manufacturable offering than Barcelona (and possbily Nehalam?), although clearly it's not going to be down to the level of Penyn's 107mm^2. Wouldn't this give AMD scope to flourish in the low end/midrange volume markets?


* Are url tags not permitted for new posters, I notice other people have access to use them?

Scientia from AMDZone said...

John

"Remember when Scientia said that AMD was fooling us with its SPEC scores and that those were simulated K8 quads and not Barcelona?"

Yes, I remember quite well. I'm baffled though why you would be so silly as to try to argue against it when it was also written up by both Ou and Kubicki two months after I did. How did you miss Ou and Kubicki's shrill protests that AMD was cheating by using simulated scores? Also, did you fail to notice that simulated was printed on the graphs themselves? If you somehow missed all of these things then I can't imagine what you did notice.

"Remember when Scientia said AMD would need ~2.8GHz to match Intel's 3.3GHz?"

Ah, you are finally making a real point. Yes, I did say that based on the spec scores. Clearly, those scores were not representative or perhaps K10 is broken. At any rate, I'm not sure at this point that a 2.8Ghz K10 could beat a 2.8Ghz from Intel.

Scientia from AMDZone said...

lex

"9) INTEL 65nm raw transistor performance at a given leakage is superior to AMD's/IBMs, TSCM and most other companies... the same part would be anywhere from 5-20% faster or lower power on the INTEL process."

You would have to give a source for this. To my knowledge, AMD has never made the same claims for 45nm that Intel has so I think it is safe to asssume that Intel is ahead with high K. And, I know that Intel has achieved a smaller gate size with their 65nm. But I have no idea where you get the 5=20% faster claim from.

"Thus if and when INTEL every starts doing graphics this advantage will overwhelm both nvidia and AMD-ati"

Well, if Intel ran off GPU's in the same way they run off CPU's then they would have the fastest GPU's through, say, mid 2009. However, I'm not sure that they could hit the same price point for the majority of the cards.

"10) INTEL raw process advantage at 45nm ... increases the fundamental power/performannce advantage by probably another 30% over the AMD 65nm process giving them a whopping > 40-50% advantage."

I think you are dreaming here. I'm certain that Intel has an advantage at 45nm, but 50%? Unlikely. Again, give sources if you can prove this.

" When AMD finally migrates to 45nm the best they can hope for without HighK / MetalG would be to be again behind by 20-30%"

Okay, so your prediction is:

Intel:
45nm versus AMD 65nm, 40-50% lower power draw.
45nm versus AMD 45nm, 20-30% lower power draw.

We'll see if this is true in Q1 and Q2. It will probably take that long for AMD to reach 3.0Ghz quad clock speeds.

"a) AMD continues to limp along falling further and futher behind... like what happened to PowerPC, Alpha, and / or Transmeta"

Unlikely. AMD would have to fall a long, long, long, long, long way to be like Transmeta.

"b) AMD somehow finds a white knight with 2-4 billion cash to burn...AMD holds grounds and catches up at 22nm."

Also unlikely. AMD will need to get its processes in order by mid 2009 (if it does high K on 45nm) or 32nm at the latest. AMD absolutely cannot wait until 22nm.

"c) AMD folds up and changes business direction in the next 6 months"

You have now drifted somewhere between delusional and schizophrenic. It would take AMD a minimum of two years to significantly change direction. Six months is impossible. Or are you trying frantically to make this fit into your absurd claim that AMD will go bankdrupt in 2008?

"d) INTEL returns to its old poor exeuction days and fails to excute 45nm ramp, 32nm development, slips nehelem, and/or both of their 32nm Tick/Tocks "

Frankly, I have no idea what time period you are talking about. And, slipping Nehalem probably wouldn't matter that much. Bulldozer is mid 2009 so anything sooner than that is good for Intel.

"e) INTEL missed something on 45nm HighK/MetalG collapses"

??? Intel has demoed 45nm. It would be more realistic at this point to speculate on a possible miss at 32nm.

"What is there to argue, AMD's business plan is broken as it is predicated on failure of your competition and not anything AMD directly controls. "

I think your basic assumption here is incorrect. AMD's plan is not based on any Intel failure. It does however require reasonably good execution on AMD's part.

"You are saying that it was design rule differences that doomed Prescott and that different design rules on Cedar Mill and Merom was what set the Core2Duo apart?"

No. I never said that. What I said was that I have seen other people claim that Intel got ahead by instituting RDR. The Prescott design was clearly bad so presumably these changes would have had to have been after Prescott.

I don't personally think there is any validity to these claims but these were the very ideas that were being swallowed without question on roborat's blog.

Scientia from AMDZone said...

The_Wolf_Who_Cried_Boy

"Appologies if this post seems somewhat pro-Intel given where I'm posting it."

The article is entirely factual; what reason would you have to apologize? Or have you been listening to the trolls who keep claiming that this is an AMD fan blog? If you want to see a real AMD fan blog take a look at Sharikou's. The truth is that even AMDZone and AMD Watcher post negative items about AMD.

The article link you posted was interesting. However, I'm somewhat puzzled because if Atomic Layer Deposition is a problem then presumably AMD could have obtained Atomic Vapor Deposition technology from Aixtron.

Scientia from AMDZone said...

wolf

You can do url tags without any trouble; you just have to use html format:

< a href = "adddress" > display text < /a >

I've added spaces so that the software wouldn't interpret the code as a literal link. Just remove the spaces except for the one between 'a' and 'href'

Scientia from AMDZone said...

I'm still a bit puzzled exactly how Intel got ahead in this area.

AMD was talking about Performance of Nitrided Hf Silicate High k Gate Dielectrics back in 2003.

And, so was IBM: Growth and Characterization of Al2O3:HfO2
Nanolaminate Films Deposited by Atomic LayerDeposition (ALD)
.

The EETimes article doesn't seem to say exactly how Intel gained an advantage. However, by all accounts, AMD does not plan to use high K in 2008. If high K is the only way to catch Intel then it won't happen until 2009 at the earliest.

The_Wolf_Who_Cried_Boy said...

My first post had an appology as I have an unfortunate habbit of not playing nice with others online and thought it might show some measure of good manners as I'm coming to your blog and posting an artical glowing of Intel's accomplishments given you do post as Scientia from AMDzone. Sorry if that itself was interpreted as antagonistic.

This is the correct link from my previous post, thankyou, tags are working fine now.

PS It's the Sydney Opera House, not that I'm in a commanding postion to lecture about spelling...

Ho Ho said...

scientia
"Well, if Intel ran off GPU's in the same way they run off CPU's then they would have the fastest GPU's through, say, mid 2009. However, I'm not sure that they could hit the same price point for the majority of the cards."

Well, NVidia sells way bigger dies than K10 for around 480mm^2 dies for $500 ($455 after rebates) and cut down versions for $300 ($260 after rebates). Why couldn't Intel match those prices if it manufactures the GPUs in-house?

Aguia said...

enumae,
Paper launched? They are on Newegg, although they were sold out rather quickly, I think about 2 days, they are there now.

Nope paper launched.
Yes it’s listed on the site but can’t be bought so that’s basically the same thing.
It appears today that there are two, one for $289.99 the other $309.99.
Also Ati 3870 seams to have been paper launched too, since it’s not listed. The 3870 costs $219.00!
The 3850 is available at just $179.00 almost half the price of one 8800GT. Good deal buying two for the people that have Intel motherboards (P965/975/P35/X38) with two slots.


When I said this it was in regards to the performance of the 3870 vs the 8800GT and I should have been more specific.

Yes performance is important has features too. Go ask the guys that bought the nvidia 6800 instead of the x800 if it was because of the performance, if it was they bought the wrong card.
And I don’t think playing a game at 30fts or 34fps makes a huge deal or 100fps vs 110fps. DX10.1 is maybe slow with today’s hardware but wasn’t the same thing Ati said when their cards missed SM3.0 in their feature set. You may eat PR talk I don’t.

enumae said...

Aguia
Yes it’s listed on the site but can’t be bought so that’s basically the same thing.


If you can add it to your cart, you can buy it, sorry you can't understand that.

And I don’t think playing a game at 30fts or 34fps makes a huge deal or 100fps vs 110fps.

Please look at HardOCP review of the 38** series, and look at the settings for actual game play, and then tell me about 4 or 10 FPS.

The 38** series has a nice price point, but performance matters more for me. The 8800GT for around $300 is great, and ATI/AMD can not compete there.

You may eat PR talk I don’t.

That is funny. Don't let someone like the developer of Crysis sway you... I mean it's not like they have the best and most graphically intense game out...LOL!!!

Aguia said...

enumae,
If you can add it to your cart, you can buy it, sorry you can't understand that.

Yes you can you are right. And you can also wait 4 week for them. It was you who said when the 8800GT was out it sold out in two days. Now when was the 8800GT released again?


Please look at HardOCP review of the 38** series, and look at the settings for actual game play, and then tell me about 4 or 10 FPS.

If I’m going to stick with just one review, and base my all point in one review maybe I’m doing something wrong.
Also this was in hardocp site:
“AMD's ATI Hits a Home Run with 3800 Series.
Great gaming at a cheap price? ATI is all over it with the 3800 series.”
Also in the hardocp review did you see the 3850 beat the 3870?! And also beat the 8800GT?! Something wrong there...
Wait it used higher detail setting on the 3870 with 1280x1024 and lower detail setting on the 8800GT with higher resolution…
LOL the card is faster than the 8800GT in Crisys...


The 38** series has a nice price point, but performance matters more for me. The 8800GT for around $300 is great, and ATI/AMD can not compete there.

So you want to pay 300$ for a card is that it? I will sell you my 8800GTS 320MB for 300$ if that’s the case. Do you want it?
And I fill sorry for you to think that price is great (since its higher than Nvidia stipulated price), I thought there was some whiners saying Nvidia cards where expensive due to lack of competition, all I see is that you guys like to pay much for the products good or bad. Simulation: “What you card just cost 200$ whoooo must suck because mine cost 300$...”


That is funny. Don't let someone like the developer of Crysis sway you... I mean it's not like they have the best and most graphically intense game out...LOL!!!

Are you talking of that game where you have to rename the file in order to some issues disappear with the Nvidia cards. Hummm…
Well if I was going to ear what developers say, then we all would be running Ati cards since the developer of the best game (Half Life 2) said their (Ati) cards are much better.

Ho Ho said...

aguia
"Yes it’s listed on the site but can’t be bought so that’s basically the same thing."

I remember people saying that K10 wasn't paperlaunched even though even 1'st tier partners can't get the chips months after the "release".


"Go ask the guys that bought the nvidia 6800 instead of the x800 if it was because of the performance, if it was they bought the wrong card."

Depending on games, NV ruled OpenGL games back then. I haven't looked it lately.

As for DX10.1, it is pretty much meanignless as it doesn't offer much. It isn't usable before Vista SP1 anyways and I wouldn't be surprised to see NV providing the features with a simple driver update


"Also in the hardocp review did you see the 3850 beat the 3870?! And also beat the 8800GT?! Something wrong there..."

Did you see the settings those cards were using? Guess not.


"Well if I was going to ear what developers say, then we all would be running Ati cards since the developer of the best game (Half Life 2) said their (Ati) cards are much better."

Would you like me to talk about how Valve royally screwed entire FX line without absolutely any good reason what so ever? I guess that $8M from ATI didn't have anything to do with it ...

AndyW35 said...

If I was running a GF6 or 7 series and or an ATi X8or X18/19 series then I would be very tempted by both 8800GT or 3850 in crossfire or single 3870. Both arrive just in time to allow good settings with high cry or very good settings with COD4 and others.

The RV670 core is so small in comparison to the HD2900 that I wonder how much margin the part has for AMD, surely a lot more.

They are the first good Ati cards since the X19 series.

Aguia said...

I remember people saying that K10 wasn't paperlaunched even though even 1'st tier partners can't get the chips months after the "release".

Wasn’t me for sure.
Besides since OEM and retailers don’t have it, what more proof do you want?
I think the Phenom will also be paper launched too. If we where seeing already lots of servers with AMD quad core CPUs I might have other opinion.


Depending on games, NV ruled OpenGL games back then. I haven't looked it lately.

Well lately the older Ati cards outperform the older Nvidia cards in the more recent games.


As for DX10.1, it is pretty much meanignless as it doesn't offer much. It isn't usable before Vista SP1 anyways and I wouldn't be surprised to see NV providing the features with a simple driver update

I don’t think so, if 8800GT was already DX10.1 Nvidia was every where announcing it. However I might agree with you, because the 8800GT is already killing all the Nvidia sales of 8600GTS/8800GTS 320/640/GTX 768, what might happen if Nvidia announced new features with the card.
I will also be interesting to see what Ati will do with the 2900XT.


Did you see the settings those cards were using? Guess not.

Did you read the rest. Guess not.


Would you like me to talk about how Valve royally screwed entire FX line without absolutely any good reason what so ever? I guess that $8M from ATI didn't have anything to do with it ...

If you are going to bring conspiracy theories then all the games that say TWIMTBP are all slowing down Ati cards.
And don’t tell me that the Nvidia FX line was good cards? They couldn’t out perform the Ati cards at any game (except the ones where they used cheats) and in all the lower image quality was clearly visible.

Ho Ho said...

aguia
"Wasn’t me for sure."

Perhaps so. IIRC Scientia was defining paperlaunch as not having parts in Newegg month after the release.


"However I might agree with you, because the 8800GT is already killing all the Nvidia sales of 8600GTS/8800GTS 320/640/GTX 768, what might happen if Nvidia announced new features with the card."

Those are EOL'ed anyway. Also if it is possible to add DX10.1 support to G92 then it will be availiable on all the older 8000 series GPUs also.



"I will also be interesting to see what Ati will do with the 2900XT."

So far I haven't heard any rumours about ATI doing a driver upgrade for 2900, I have for NV.


"If you are going to bring conspiracy theories then all the games that say TWIMTBP are all slowing down Ati cards."

No conspiracy theories. I tried out myself and forcing DX9.0 shaders with 16bit accuarycy on my FX didn't have any cisual defects but ran twice as fast as with 32bit accuaricy.


"And don’t tell me that the Nvidia FX line was good cards?"

They weren't as good but still Valve crippled them without a good reason making the difference look even greater.

Scientia from AMDZone said...

wolf

As I said, there is no need to apologize for posting something factual that is negative about AMD or positive about Intel. The reason my name is Scientia From AMDZone is so that people know I'm the same Scientia who posts there (I am not the only Scientia on the internet).

If you are confused about the difference between good quality posts and poor posts I suppose we could go over one of Lex's posts as a critique.

With 45nm arrival and a good Penrym and Nehalem Tick-Tock intel has put togather superior silicon with huge manufacturing capacity and is now using that to manufacture a good design.

Good paragraph. Concise, factual, and makes its point clearly.

Good design + great silicon + huge manufacturing is competing against Good design + poor silicon + small capacity

The first part is simply repeating the previous paragraph. The second is not really accurate.
Good design - I don't think we could infer this yet. The best we could infer is that K10 might be a good design.
poor silicon - Okay, this could be inferred from the low clock speeds.
small capacity - Small here is very misleading. AMD cranks out over a million chips a week. AMD has about 30% of Intel's capacity but it hasn't topped out FAB 36 or converted FAB 30. It has been suggested that AMD would be viable at 40-50% of Intel's processor production capacity. This would mean that Intel would still be making twice as many processors as AMD so by Lex's scale this would still be "small". For comparison, Anheiser Busch has a similar share of the beer market while Ford has about the same share of the medium duty truck market. As far as I know, neither Miller nor General Motors have decided to vacate.

Oh... its in a business that takes billions of dollars to play...

This statement is simply trivial. AMD spent about $2 Billion building FAB 36 and a similar amount buying tooling. AMD has surpassed previous heavy hitters like Motorola whose process tech was first rate nine years ago.

Anyone want to predict the winner and the loser?

This sentence is a minor troll. The statement adds nothing factual nor composes any elements into a conclusion. It's basically just a childish provocation.

enumae said...

Aguia
Yes you can you are right. And you can also wait 4 week for them.


I can order it now, with 2 day shipping.

LOL the card is faster than the 8800GT in Crisys...

This is an example that proves you can't or don't want to understand what is being shown in the review. What is harder to do, run a game at 1280x1024 or 1600x1200?

So you want to pay 300$ for a card is that it?...

The statement I made was simple, ATI/AMD do not offer the performance to sell there products at $300. Can you really not understand that?

Are you talking of that game...

Show me someone besides AMD that is claiming that DX10.1 is important to any currently or soon to be released games.

Aguia said...

This is an example that proves you can't or don't want to understand what is being shown in the review. What is harder to do, run a game at 1280x1024 or 1600x1200?

Go back to hardopc and read the detail setting again and answer me this:

What is harder?
1280x1024 with
Medium/Medium/High/High

OR

1600x1200 with
Medium/Medium/Medium/High

I'm starting to think it’s you who don’t want to understand.


The statement I made was simple, ATI/AMD do not offer the performance to sell there products at $300. Can you really not understand that?

And how is that bad?
Less 100$ for less 10% in performance; easily compensated with more features, less noise and less heat.


Show me someone besides AMD that is claiming that DX10.1 is important to any currently or soon to be released games.

The past tell me that, which is the most important thing.
Want to bet there will be games with features that can only be enabled if you have a DX10.1 (SM4.1)?
And want to bet with me there are games that could be perfectly run on the X800 cards but don’t because they lack SM3.0?
Want to bet with me as soon Microsoft releases Vista SP1 one of the highlighted feature will be the DX10.1?

enumae said...

Aguia
I'm starting to think it’s you who don’t want to understand.


1. 1280x1024 = 1.25MP

2. 1600x1200 = 1.83MP

The resolution of 1600x1200 is 1.4 times or 40% larger than the 1280x1024.

So again, what is harder to do, run a game at 1280x1024 or 1600x1200?

And how is that bad? ...

I can see that you don't understand the point so I will drop it.

The past tell me that...

Again, show me someone besides AMD that is claiming that DX10.1 is important to any current or soon to be released games.

MaxR said...

Again, show me someone besides AMD that is claiming that DX10.1 is important to any current or soon to be released games. Check this out.
http://www.theinquirer.net/gb/inquirer/news/2007/11/16/why-dx10-matters

enumae said...

MaxR

Thanks for the link. I have seen that and in all honesty if AMD were to fart Charlie Demerjian would say it smells like roses :).

In regards to "multi-sample buffer reads and writes", I wish I could say I understand it, but in having read the article I do see where it says...

In the end, DX10.1 is mostly fluff with an 800-pound gorilla hiding among the short cropped grass. MSBRW will enable AA and deferred shading, so you can have speed and beauty at the same time, not a bad trade-off.

Now while it may be a nice feature and have great promise I still have the question.

When will or what games will use this?

Currently the best games don't talk about it, Crytek went so far to say DX10.1 isn't worth it and current hardware isn't fast enough to use it.

All in all Aguia wants to be proud of an AMD/ATI video card by any means possible including AMD/ATI having DX10.1 even though you can't use it.

Ho Ho said...

enumae
"When will or what games will use this? "

Currently probably only Stalker could benefit from it but not more than <5%. Having the ability to read Z-buffer directly without having to write it to a texture yourself doesn't really help all that much. There was a lengthy discussion on beyond3d about that where a few game developers talked about it.

Aguia said...

1. 1280x1024 = 1.25MP
2. 1600x1200 = 1.83MP
The resolution of 1600x1200 is 1.4 times or 40% larger than the 1280x1024.
So again, what is harder to do, run a game at 1280x1024 or 1600x1200?


I can see that you don't understand the point so I will drop it.
I didn’t know that screen resolution was the only in game detail setting...


I can see that you don't understand the point so I will drop it.

If I where you I would be hard to understand too.
Since I see you have your wallet full...


Again, show me someone besides AMD that is claiming that DX10.1 is important to any current or soon to be released games.

Since nobody had it its not important. You see SSE4 for example is also insignificant since nobody besides Intel talks about it...

Aguia said...

All in all Aguia wants to be proud of an AMD/ATI video card by any means possible including AMD/ATI having DX10.1 even though you can't use it.

Tell me:
Weren’t you one of the guys that was bashing Ati for not having released their DX10 part sooner?
Weren’t you one of the guys that prised DX10 support on the Intel 945?
And bashed Ati X1x50 chipset because it didn’t?

I think you I’m already seeing you writing something like this if the pages flipped:
“Again Ati is late, Nvidia is already releasing DX10.1 capable cards ahead of time, and even Microsoft cant keep up with the Nvidia advanced design cycle.”


And this is from you right?
“4. Delays for R600 based DX10 graphics, while Nvidia is launching there mid to low end cards, and regardless of their drivers, they have shipping products.”

Aguia said...

ho ho and enumae,
There is also something called longevity of the card:

legion
anand

Tell me where you see the older nvidia cards and the older ati cards in those charts?
What conclusion is also taken from the legion hardware review regarding the CPU?

And what to conclude from this “updated” test:
legion
Not bad for a “broken” card with “slow” clocked shaders.

Ho Ho said...

aguia, can you find a single site showing 8800GT being beaten in Crysis by the new HD GPU?

Also I repeat again, DX10.1 is rather insignificant update to DX10. Most gamedevs have said that, only AMD has advertised it as best thing ever. No wonder, it has much else that would be better with its GPUs.

as for Legion benchmarks, this is what they said:

"The GeForce 8800 GTX appeared to be more consistent, with no real drastic fluctuations in frame rates. The Radeon HD 2900XT seemed to drop a lot more frames when walking into bigger more open rooms."

Of course as R500 in XB360 and R600 and its derivates are rather similar in nature there is no wonder that Bioshock is better optimized for it

Ho Ho said...

Woha, lots of typos in my last post. Sorry.

beaten in Crysis by the new HD GPU? at the same settings

No wonder, it doesn't have much else that would be better with its GPUs."

Aguia said...

Here is an excellent review that matches my all points:

Gamespot

Nvidia 6800 VS Ati X800
9 fps VS 30 fps

Ati old card wins by 3.3X.
Even the newer 7900GS is beaten by a much older generation card.

*****

Nvidia 7600GT VS Ati X1650XT
16 fps VS 32 fps

Ati old card wins by 2X.
7900GS from another level is also beaten by 33%.

*****

Nvidia 7900GS VS Ati X1950 Pro
17 fps VS 33 fps

Ati old card wins by 2X

*****

Nvidia 7900GTX VS Ati X1900XTX
17 fps VS 30 fps

Ati old card wins again by almost 2X
Even the slower X1950PRO beats the Nvidia card by 50%!

*****

Nvidia 8800GTS 320 VS Ati X1950XT

17 fps VS 17 fps

Ati old card matches a brand new Nvidia card in performance. Amazing!


That also explains the profits Nvidia have over Ati, the Nvidia guys have to upgrade much more ;)

enumae said...

Aguia
I can see that you don't understand the point so I will drop it.


"The GeForce 8800 GT was better performing than the 3870 as the 8800 GT was able to game in Crysis at 1600x1200 with quality settings on “High” except shadows, post processing, and shader quality."

That is quoted from the review...

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

You see SSE4 for example is also insignificant since nobody besides Intel talks about it...

The differencec is that if you do Video encoding and have a Penryn processor you can actually use SSE4.

ATI/AMD may have DX10.1, but you can not use it. See the differeence?

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

Weren’t you one of the guys that was bashing Ati for not having released their DX10 part sooner?

I don't know if I was bashing, The discussion was about ATI/AMD financials and how they needed to release it's competing product.

I will admit that I have not gotten a new video card since my X1900XT because I am hoping for ATI/AMD to have a new and better product, I like ATI GPU's and don't want to use Nvidia and so I now find myself waiting for the 3870X2.

Weren’t you one of the guys that prised DX10 support on the Intel 945?

I don't think so, if I had I would have been incorrect.

And bashed Ati X1x50 chipset because it didn’t?

Again I don't think so.

Are you talking more about the ability to run Aero on the 945 chipset?

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

And this is from you right?

Yes, and it was breaking down AMD's current events as of April 21, 2007.

Was I lying?

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

There is also something called longevity of the card:

Is that like 64bit from AMD, just kidding.

I understand what your saying but you have gone from DX10.1 is good because ATI/AMD says so to the longevity of the GPU...

You are just trying to make your point without understanding that currently, and near future DX10.1 is and may not be important.

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

Here is an excellent review that matches my all points:

You are funny!!!

Aguia said...

Enumae,
That is quoted from the review...

"The GeForce 8800 GT was better performing than the 3870 as the 8800 GT was able to game in Crysis at 1600x1200 with quality settings on “High” except shadows, post processing, and shader quality."

But according to them the Ati 3870 was able to game in Crysis at 1280x1024 with quality settings on “High” except shadows and shader quality.

Now what slows down more the game, post processing or resolution?
Do you know the answer?
Then tell me because I surely don’t!


ATI/AMD may have DX10.1, but you can not use it. See the differeence?

No difference at all!
Because the encoding you are talking about just works with some specific codec’s not all.
Funny that DX10 also has the same issues I just described has will the DX10.1 will not benefit every game. I think ho ho was too nice saying <5% more like <2% in the first year.

And I may not use the technology today put I’m sure that I will use it later and that my hardware will be ready for it.


I don't think so, if I had I would have been incorrect.

Well you are out of luck because I have found previous discussions where you prised it. So if you now admit you where wrong than at least is a good start ;)

Again I don't think so.

I put the link or do some copy paste…


Are you talking more about the ability to run Aero on the 945 chipset?

No. But there was some discussion going on with that.

Was I lying?

No. But you put DX10 on the sentence so you wanted DX10.


I understand what your saying but you have gone from DX10.1 is good because ATI/AMD says so to the longevity of the GPU...

No. It has nothing to do with DX10.1.
Longevity by performance! Ati old cards still perform extremely well in new games and Nvidia cards "suck". Period!


You are funny!!!

Really?! Ati winning all the tests VS Nvidia by a 2X difference? Maybe its funny for the ones that bought Ati cards but for the guys that bought some Nvidia card I don’t think so.


I will admit that I have not gotten a new video card since my X1900XT because I am hoping for ATI/AMD to have a new and better product, I like ATI GPU's and don't want to use Nvidia and so I now find myself waiting for the 3870X2.

Maybe you can give it to me ;) mine 6600GT (AGP) is already showing its age...

enumae said...

Aguia
Then tell me because I surely don’t!


I have tried. The 8800 running at 1600x1200 (an increase of 40%) vs 1280x1024 while being 2 or 3 FPS slower than the 3870 makes the 8800 faster. For any other comparisons look at the other test in the review.

No difference at all!

Ok, so being able to use something today (while limited) vs ? (who knows when) is no different?

Well you are out of luck because I have found previous discussions where you prised it.

Please link the discussions.

I don't know why I would say the 945 chipset supports DX10, the discussions in the past about the 945 were in regards to Vista and Aero.

But you put DX10 on the sentence so you wanted DX10.

Was I lying... No! DX10 capable cards are capable of running older games faster due to the better hardware, so yes I wanted a DX10 card for the improved hardware and effect it couls have on my gaming (BF2).

Maybe you can give it to me ;) mine 6600GT (AGP) is already showing its age...

Well it will be on Ebay after the holidays (if youd like I will post a link).

Ho Ho said...

aguia
"Now what slows down more the game, post processing or resolution?
Do you know the answer?
Then tell me because I surely don’t!"


Depending on game but for Crysis it is surely resolution as its shaders are rather complex


"I think ho ho was too nice saying <5% more like <2% in the first year."

I meant that if a game would really use all that DX10.1 offers it could be around 5% faster, not that 5% games would benefit from DX10.1. Also it is worth considering that vast majority of DX10 level HW owners do not run Vista. Abother point is that NV has around 60-70% marketshare. I doubt there would be many gamedevs who would bother supporting a feature if it is working only on a very few PCs. Of course if rumours are correct and 8xxx series do get 10.1 features with a driver upgrade then things could be different.


"And I may not use the technology today put I’m sure that I will use it later and that my hardware will be ready for it."

Well, by the time real DX10.1 games are out that GPU will be rather old and not fast enough to really enjoy it. Same as with first DX9.0c HW.


"Longevity by performance! Ati old cards still perform extremely well in new games and Nvidia cards "suck". Period!"

The difference between ATI and NV during last few generations have been that ATI HW targets stuff that is yet to see the light of day whereas NV targets things that are there by GPU launch. Basically what that means is that ATI crams a whole lot of shader power but leaves the rest to be desired. Of course once there actually exist games that are heavy on shaders ATI old HW looks quite nice, even though I wouldn't call 800x600 with medium quality at <20FPS exactly nice gaming experience. See my point about HW getting dated as time passes.

xwraith said...

Hey Scientia:

Here is an article from the inquirer that may shed some light on why things aren't performing up to expectations:

http://www.theinquirer.net/gb/inquirer/news/2007/11/18/amd-delays-phenom-ghz-due-tlb

Thoughts?

Scientia from AMDZone said...

Ho Ho

"IIRC Scientia was defining paperlaunch as not having parts in Newegg month after the release."

Sorry, I keep forgetting how young you are. The definition is not mine and is not new. It comes from Anand Lal Shimpi from seven years ago.

Intel Strikes Back - Spring IDF - February 2000

The difference between AMD's launch and Intel's that took place no more than two days later was tremendous. For the first time in quite a while, AMD's processor was going to be available, in mass quantities before Intel's. The 1GHz Athlon as well as the 900 and 950MHz parts that accompanied it would all be available by the end of the month in March. Intel however would not push for the release of their 850, 866 or 933MHz parts in spite of the fact that they had already released the 1GHz Pentium III, and even then the 1GHz Pentium III was released in "limited quantities." It wouldn't be until another quarter later that the Pentium III would be available in greater numbers.

This was truly the birth of what Intel became notorious for in Y2K, the Intel "paper-launch". It was obvious that the early release of the 1GHz Pentium III was no more than a marketing tactic, a paper-launch to say: "we have a 1GHz CPU too." It is interesting to note that just one year prior to this AMD would have been listed as the paper-launching CPU manufacturer, while it would have been unheard of that Intel would do such a thing.

Intel’s 1.13GHz CPU Recalled – Is Intel resorting to desperate measures?

Since the release of AMD’s Athlon, things have changed. Slowly but surely the roles of the two companies have reversed, now, Intel is the one being accused of “paper launching” processors while AMD CPUs are readily available and definitely affordable. These “paper launches” were at their worst with the release of the 1GHz Pentium III (March 2000) before the 850, 866 and 933MHz Pentium IIIs in an attempt to compete with AMD’s 1GHz Athlon that was released just days before. What began to make the community characterize Intel’s CPU releases as “paper launches” was the fact that you couldn’t actually go out and buy a 1GHz Pentium III whereas, by the end of the month, the Athlon was already available in speeds from 500MHz up to 1GHz in 50MHz increments.

The trend seemed to continue as Intel released the 1.13GHz Pentium III at the very end of July, while the 1GHz Pentium III parts were just finally becoming available in decent quantities, four months after the “release.”

*******

I think this is still a good definition. So, if you want to characterize a processor launch then just reference these two articles for both AMD and Intel.

Scientia from AMDZone said...

To me and going by Anand's comments I would have to say that a few tens of thousands of processors for Barcelona was a pretty soft launch. Consider that having half a quarter's worth of processors at 5% production would be over 300,000 procesors. But AMD's launch was only a fraction of this.

Giant said...

Hexus PHENOM review:

http://www.hexus.net/content/item.php?item=10427&page=1

See not only the awful performance of PHENOM vs. C2Q Q6600 but also the substandard performance of AMD's chipsests in HDD performance, USB performance etc. vs. Intel's chipsets.

Also, some excellent quotes from the article:


Irrespective of whether you think that Intel's glue-dual-cores-together approach is architecturally inelegant, the fact remains that Core 2 Quad - in both its Kentsfield and new-and-improved Penryn flavours - is a fast and efficient processor in practically every way.




AMD's nascent Phenom also suffers under the considerable yoke of Intel's Core 2 Quad 6600 pricing, which at £165 for a hugely-overclockable 2.4GHz part is something of a bargain. AMD, though, is pitching its slightly underperforming quad-core part at roughly the same price. The industry needs AMD to survive and succeed yet it's very difficult to make a compelling buying recommendation for a processor that's a year behind its competitor - one who has already moved on to a more-efficient 45nm manufacturing process - is between 10-20 percent slower in most benchmarks, and costs much the same.


. Right now, pressed for buying advice, we'd recommend our readers opt for the competition's processor, chipset, and graphics cards.

AMD is a joke right now! All these 'next generation' products (Barcelona, AMD 790 chipset, Phenom CPU, ATI 3800 video cards are ALL slower than the competition's existing parts!)

Hexus even took pity on AMD: Using Intel's slowest quad core CPU for a comparison. It would have been nice to see the new QX9650 in there too.

Axel said...

Giant

Hexus PHENOM review

Confirmed to be slower than Kentsfield per clock, as the illicit previews all year have shown. Even worse, the 2.4 GHz Phenom 9700 has apparently been delayed to late Q1 08.

Here are a couple other reviews confirming this cold hard truth:
Hot Hardware
Tom's Hardware

Meyer will inherit a trainwreck of massive proportions from Ruiz.

Mo said...

Hexus, Hot Hardware and Tom's Hardware....all put the Q6600 at LESS POWER USAGE while outperforming the 9700 Phenom.

Phenom 5-10% slower than Core in Clock for CLock comparision.

Sure... call the reviews biased or paid intel pumpers but then you're saying that EVERY reviewer out there is paid...........Hardly.

Sci, can you elaborate on what went wrong for AMD? and what can be done now.

T800 said...

I've only had a chance to skim through the reviews, but so far it is looking a bit depressing for AMD. I did not have ultra high expectations, but I was hoping for AMD to come out swinging and trade some blows with Intel.

On another note, because of the problems with the AMDZone.com forums, I've setup my own forums. They're fresh off the presses so hopefully things will run smoothly. Scientia if you consider this spam then feel free to remove.

www.amdmatrix.com/forums

Aguia said...

Scientia,

Judjing but the Anand Intel's Core 2 Extreme QX9770 Preview, Intel is having issues again similar to the Pentium 1.133Mhz you just described.

Scientia Anand is also talking of this
“AMD couldn't simply get enough quantities of the Phenom at 2.4GHz to have a sizable launch this year (not to mention a late discovery of a TLB error in the chips),
Where isn’t too explicit if the bug still exists or not and the consequences of it.


Giant,

. Right now, pressed for buying advice, we'd recommend our readers opt for the competition's processor, chipset, and graphics cards.

Are they dumb or what. Since when you base the all spectrum of products, in just one specific product, with one price. All the Ati products 19x0,2400,2600, 2900, 38x0 beat all the Nvidia products up to 300$.


Hexus even took pity on AMD: Using Intel's slowest quad core CPU for a comparison. It would have been nice to see the new QX9650 in there too.

I would compare products that cost the same.

Ho Ho said...

aguia
"Judjing but the Anand Intel's Core 2 Extreme QX9770 Preview, Intel is having issues again similar to the Pentium 1.133Mhz you just described."


That QX9670 was an ES chip and they are know to be very with varying quality. Just read the phenom reviews, People reported that there was a special box for OC testing as the ones given to journalists simply didn't OC almost at all. Let's wait and see what happens when the chips hit retail.

People didn't want to believe the pre-release Phenom benchmarks, why not do the same with intel CPUs?


"All the Ati products 19x0,2400,2600, 2900, 38x0 beat all the Nvidia products up to 300$."

On average or in some specific benches? I'd like to see what GPU beats 8800GT.


"I would compare products that cost the same."

Then we should be comparing Q6600 vs 9500. On average Q6600 was about 15% faster. Would that make you happier than comparing clock-to-clock?

sharikouisallwaysright said...

38xx is a huge succes, it has the right price for more than enough performance and will atract the majority of the market if only available.
Let Nvidia sell some tenthousend highendchips, who cares, the money comes from the masses!

Phenom is a rough launch, but what matters isnt the start but the result in the end.
I expect better performance from higher clocked chips AMD has to deliver, otherwise they will fail.

Its far to early to jugde the new chipsets.

Ho Ho said...

"I expect better performance from higher clocked chips AMD has to deliver, otherwise they will fail."

Without superlinear scaling and extremely fast clock speed ramp it would be rather difficult.

Aguia said...

ho ho

People didn't want to believe the pre-release Phenom benchmarks, why not do the same with intel CPUs?

You are right ho ho. But then again why would Intel send them one bad CPU that makes them look bad when they didn’t need it?


On average or in some specific benches? I'd like to see what GPU beats 8800GT.

That’s why I said under 300$. When the Nvidia 8800GT costs less than 300$ I will update my base line ;)


Would that make you happier than comparing clock-to-clock?

Yes it would. Price is a very good measure.
If AMD is slower and it seems it is, then Intel have won this round and no more cheaper Intel Core 2 Quad. I was expecting AMD could at least keep up so it would force Intel to lower the prices. It seems they will not.
Remains to be seam if AMD will lower their prices. 200$ would look good.

Axel said...

So out of curiosity, I wonder if Rahul Sood of HP/VoodooPC knowingly lied when a couple months ago he spouted the nonsense that Phenom 3.0 Ghz would "kick the living crap" out of any CPU then on the market. Or did he unknowingly regurgitate the lies told him by someone at AMD? Either way, I'm now crossing Sood off my list of credible sources. Good riddance!

13ringinheat said...

Aguia says:That’s why I said under 300$. When the Nvidia 8800GT costs less than 300$ I will update my base line ;)

Its time for an update Best buy selling 8800GT for $249......go look at this weeks flyer

Ho Ho said...

aguia
"But then again why would Intel send them one bad CPU that makes them look bad when they didn’t need it?"

It is bad in terms of thermals, not speed. Thermals have often been quite bad for ES chips. Also I remind you the fact that even at 3.6GHz the earlier Penryn chips didn't use as much power as the new ES one anand had.


"When the Nvidia 8800GT costs less than 300$ I will update my base line ;)"

How about now? Sure, it is out of stock at the moment but price is certainly <$300


"Remains to be seam if AMD will lower their prices. 200$ would look good."

As a consumer I'd love that. Though looking to the future that would basically be a financial suicide.


axel
"So out of curiosity, I wonder if Rahul Sood of HP/VoodooPC knowingly lied when a couple months ago he spouted the nonsense that Phenom 3.0 Ghz would "kick the living crap" out of any CPU then on the market."

He could have been comparing it to AMD dualcores. Though my guess is he just wrote what AMD had told him and took their every word as gospel.

Axel said...

Ho Ho

He could have been comparing it to AMD dualcores.

No, Sood's exact words of FUD were:

"And for the record, if you were to benchmark Phenom at 3GHz you would see that it kicks the living crap out of any current AMD or Intel processor—it is a stone cold killer."

Ho Ho said...

Well, perhaps it kills with heat when you OC it to 3GHz and it starts consuming 140W. I wonder how much would it be in Intel measurement.


"When it comes to power consumption the 140W Phenom 9900 likes to consume a lot of energy At full load it topped the charts at 345W and at idle it ran 197W with Cool'N'Quiet enabled and 243W if it was disabled. The new Intel QX9770 for example is has a higher clock frequency and more cache, but runs nearly 100W lower at idle"

Mo said...

Wow. the 9900 is one major power sucker if it is indeed 140W.

Ho Ho said...

). It's the darling of the overclocking community, so we hope to see similar results with the 2.2GHz and 2.3GHz Phenoms. From there, the TDP skyrockets. 2.4GHz pushes you up into the same thermal range as Intel's current Extreme Edition CPUs. The 2.6GHz systems AMD set up for us have a processor TDP of 140W.


So, 2.6GHz with 140W late Q1. Is it even possible to get 3GHz quad out before next Christmas with half-decent thermals? How many revisions has K10 had by late Q1? Surely it should be enough to fix most bugs there exist. If not then I'm afraid only savior AMD can hope is Bulldozer

Axel said...

Ho Ho

If not then I'm afraid only savior AMD can hope is Bulldozer

To be blunt, AMD have almost certainly known since before this past July that K10 would simply be a stopgap to Bulldozer. During the July Analyst Day conference, AMD essentially downplayed K10 and talked up Bulldozer as the real deal, probably for the sole purpose of winning over gullible investors like the Abu Dhabi government so that AMD can survive until late 2009 when Bulldozer is slated to launch.

It hardly matters anymore if 65-nm K10 can hit 3.0 Ghz by Q2 08 with decent thermals. Its IPC to die size ratio is so far behind that of Yorkfield, and Yorkfield's pricing so aggressive, that AMD will not be able to make enough money on K10 to stop the bleeding even if they can get faster clocks out by Q2. At this point we're looking at 2.4 GHz and hopefully 2.6 GHz by late Q1. I doubt that 2.8 GHz would emerge until Q2 at earliest.

Shanghai in late 2008 would bring the die down to competitive sizing but the big question is whether AMD can bring the leakage down to reasonable levels to clock them competitively.

Effectively, AMD are screwed for 2008 with their current form of operations. They must either raise a lot more cash or start selling assets. The Abu Dhabi windfall buys them a quarter or two of time to consolidate and execute on asset lite.

Aguia said...

ho ho,

Well, perhaps it kills with heat when you OC it to 3GHz and it starts consuming 140W. I wonder how much would it be in Intel measurement.

Wooo. That’s a lot.

But Intel cant laugh too much since the new Intel's Core 2 Extreme QX9770 Preview: Too Hot for TV at Anandtech.

Hello Global Warming

We were curious to see if testing power consumption would confirm our suspicions of the QX9770 running significantly hotter than the QX9650. See for yourself:


That’s 58W over QX6550 at idle and 75W more at full load over QX6550.

What just 200Mhz did, hem?
Not even the super 45nm cmos process seams to help there.

Anand Intel's Core 2 Extreme QX9770 Preview: Too Hot for TV

Ho Ho said...

aguia
"But Intel cant laugh too much since the new Intel's Core 2 Extreme QX9770 Preview: Too Hot for TV at Anandtech."

Remember, that is an ES CPU. Intel is selling 80W 3GHz quadcore Xeons. Rising frequency by 200MHz will never ever drive it that high on retail CPUs. How high is is the TDP listed on retail QX9770?

That AMD CPU is a retail one coming to sale in three months. That means it is far enough to enable a respin but AMD still doesn't believe it could bring thermals to tolerable levels.

lex said...

Scientia, I'm truely dissapointed in you!

You appear so logical but......................

You are so critical of my posts and delete my postings because of your sensitivity to my insults or propensity to repeat myself. I think you should be concerned less in my habit of repeating something. I find it humurious to see how you allow rants here about silly debates about benchmarks, power, far more silly and opinion driven then data driven.

I always talk about easily quantified and undebatable elements and or trends. Its obvious what the situation is and what the final outcome is. Debates about benchmarks and application are so specific that they simply aren't meanfuly or usefully in discussion the larger situation. You folks need to step back and stop arguing about the trees and look at the forest, it might be interesting what you see. You often don't like the extrapolations I make. I think of them as similar trends as a debate about whether stocks are better then bonds as a long term investment. Sure during short time periods stocks can look horrible, but when you step back in general there is no better investment.

Lets look at two of your comments here.

"poor silicon - Okay, this could be inferred from the low clock speeds"

or

"Good design - I don't think we could infer this yet. The best we could infer is that K10 might be a good design."


From what I have read the architecture approach that AMD took on Barcelona is excellent. Now whether they executed it cleanly is another question. Higher level decisions about clock managment, global power decisions, interfaces etc., new instruction acclerations etc... are all excellent.
Whether they allocated enough chip space for each functional block, enabled enough routing with their design rules, designed in enough timing margin with their process capability is a seperate issue. Its an open question as to whether the archticture vision and execution is aligned with the silicon capability they have currently. The fact that it is slow and/or late is an execution and planning issue related to AMD's sorry state, one I've pointed out but its repetitive and thus not relevant to discuss. THe fact I think highly of Barcelona isn't at all a postive view of allowing this architecture to go forward. It simply is the wrong architecture for AMD since their silicon capability sucks. If AMD was on 45nm and really only 3-6 months behind then it'd be okay as they would have quickly migrated the architecture to the silicon process node it needs to be own for successfuly. Barcelona on 65nm is a joke, its like the original Willaimette Netbust architecture on 180nm.

I think there was an interesting interview of INTEL's CTO who talked about INTEL's reason for not doing CSI or integrated memory controler. His answer was alone the lines that it wasn't that it wasn't something they didn't think of it was a concious decision of aligning need with silicon and business sense. In my minde both CSI and IMC are better but at somepoint it makes business sense and at another point its a lunacy. AMD is operating in the looney world without a care for sound business strategy. Forget making money, forget aligning designs, costs to be competitive with the silicon you have, forget it just make it, lose billions and blame INTEL for it all. Or maybe sucker some Arab to drop a few billion to keep you afloat.

Let me move on to your second huge fallacy, that low clock speeds mean slow or bad silicon. If you believe this you'd infer that Cedar Mill and Prescott high clock speeds say great things about INTEL silicon. But we saw already that they didn't mean anything but a bad design. On both 90 and 65nm great products came out; Core and CoreDuo on the very same process that produced the hot Netbust, both much slower in frequency but much better. The only placer where from pure frequency thing that can tell you a bit about silicon technology would be ring-oscillator, and even that can even be misleading. What you completely miss is that the capability of the silicon is the most un-ambigious thing. Every year at IEDM or VLSI the world gathers to hear about the worlds leading breakthrus and various companies silicon capability. Each presentation shows their technologies respective generation and shows clear things like SRAM size, minimum gate length, Metal pitch, and things like transistor strength and leakage. From these papers real technologists can easily determine who is ahead and by how much. I really encourage you to research a bit on the silicon piece and perhaps then you might finaly gain some real prospective of the situation instead of just deleting another post.

Greg said...

Lex, how did I know that the really long post at the bottom of the page that just complains about how scientia reads his blog would be yours...

The Phenom launch does suck quite a bit. My only question is whether this TLB bug is actually a problem across their entire product line that simply becomes uncontrollable above 2.4 ghz or maybe a cover for some other issue effecting the processors. You'll notice that some of the reviews show a serious drop in memory bandwidth once the frequency begins exceeding that number.

I see a glimpse of what could be hope in a respin of the product before 45nm with this errata fixed, possibly yielding an actual performance per core per clock increase over k8 (right now there isn't one, which really doesn't make any sense). However, this would depend on my theory being true (thus it's a very thin sliver of hope).

Of course, axel, lex, and giant are all really happy they were right about the performance of the processor, despite the fact that this means we may be losing Intels only motivation to innovate.

Greg said...

Correction, "how scientia *maintains* his blog".

I really have no idea where "reads" came from.

Giant said...

Abinstein

Please recall some of your earlier statements, regarding Phenom vs. Core 2 Quad (both Kentsfield and Yorkfield):
#1

At the same clock rate, Phenom X4 will be faster than quad-core Penryn...

#2

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.


There are many reviews on the internet. Anandtech, HardOCP, Hexus etc. all have reviews. They all agree that a 2.3Ghz is about 10% slower than a 2.4Ghz C2Q Q6600.

Looking at the facts here, and your earlier statements, it certainly appears to me that you were mistaken.

Giant said...

Obviously when I wrote about '2.3Ghz' I was referring to a 2.3Ghz Phenom CPU.

Ho Ho said...

greg
"My only question is whether this TLB bug is actually a problem across their entire product line that simply becomes uncontrollable above 2.4 ghz or maybe a cover for some other issue effecting the processors."

My guess is it is just a good reason for them to delay the introduction of faster CPUs. Come on, 2.6GHz at 140W TDP by the end of Q1 and that should be together with the TLB bug fix respin that gets done by early Q1 with 2.4GHz/125W CPU?! They are simply trying to find things they can blame so people won't wonder why it takes them so long to crank up clock speeds.


"You'll notice that some of the reviews show a serious drop in memory bandwidth once the frequency begins exceeding that number."

Any specific links? I didn't notice that in the reviews I saw. Sure, split MC wreaks havoc in most memory bandwidth benchmarks but that doesn't mean anything in real world.


"Of course, axel, lex, and giant are all really happy they were right about the performance of the processor, despite the fact that this means we may be losing Intels only motivation to innovate."

I'm not sure about them but I'm very dissapointed in AMD. I was hoping much more from them and I'm directly blaming them if it slows down industry progress. They were talking about releasing K10 at 2.6GHz at first, quickly changed that to few hundred MHz lower, paperlaunched at 2GHz and now cannot get more than 600MHz added with extreme thermals in half a year and are looking for things to blame in their problems.

Pop Catalin Sever said...

Congrats Scientia,
you were right on spot about Intel's capability to deliver Penryn at above 3.2 GHz (I don't know on what you based your assumptions on, but you were right on spot).

Also a 40% system power consumption increase for QX9770 over QX9650 is definitely a roadblock for Intel on delivering higher clocked Penryns.

I'm curious how long will it take for AMD to fix the issues with K10 and if AMD will tune the micro architecture to fix both issues and improve performance. I'm guessing we should see some improvements starting from the next stepping but right now I'm pretty much clueless what the overall effect might be.

Aguia said...

Remember, that is an ES CPU.

OK ho ho.
But then tell me there are guys on the web OC their current QX9650 to 4.0Ghz. Hey toms have one over clocked to 3.33Ghz with 6W power consuming increase. Tell me how an ES sample is much worse than current market products. At least should be about equally (performed as one QX9650 clocked at 3.2Ghz).

Is there some problem with Intel 45nm products?
Why did they increase the TDP to 136W with the QX9770?
Is everyone reading the processor power consuming incorrectly like I already asked many times here?
Are applications like prime really loading the all cores to fully extent?

Ho Ho said...

aguia
"But then tell me there are guys on the web OC their current QX9650 to 4.0Ghz."

Theirs are not ES chips. I remember old ES Core2's also consumed awfully lot of power.


"Is there some problem with Intel 45nm products?"

From what it looks like, no there is not. Otherwise Intel wouldn't be selling their 80W quads.


"Why did they increase the TDP to 136W with the QX9770?"

Beats me. Perhaps they want to have more high-end chips fitting that speedbin as the best ones are probably going to be sold as Xeons.


"Is everyone reading the processor power consuming incorrectly like I already asked many times here?"

How would be the correct way? Isn't measuring at the wall socket correct when all you do is replace the CPU? Should be good enough to see the difference between chips.


"Are applications like prime really loading the all cores to fully extent?"

Scientia said that it is reccommended so I guess it is. If prime still doesn't then what does?

Pop Catalin Sever said...

chuckula,

a little less drama will be fine.

I agree with one think that HR is not driving the company in the right direction. But that's just my opinion an the reasons are not the same as yours.

Anyway too much drama here, it's not like Intel didn't promise 10GHz or thousands of cores, or their slides were always perfectly alright (on the contrary).

What AMD promised and what it delivers only affects those that made decisions based on those promises, namely some of AMD's partners, the rest of the world that waited for the real product to show up are pretty much untouched by AMD's inability to deliver up what they anticipated.

Pop Catalin Sever said...

"Douglas McIntyre, of the 24/7 Wall Street blog, asserts that AMD needs to find a buyer - and that Nvidia (NVDA) “may be only one company that makes sense"

source

No comment, this is getting to hilarious, the irony !

Spaztic Pizza said...

I think the obvious answer to Scientia's original question for this particular post, given the evidence of the last few days is....

Yes. Yes it has.

Giant said...

NVDA buying AMD? Not a chance. Even if NVDA wanted to, the FTC wouldn't allow it, as it would give NVDA a monopoly in the discrete GPU market. If they were going to do that they'd need a seperate buyer for the graphics division.

There's also the slight fact that AMD's x86 license doesn't transfer with a purchase of AMD, it gets revoked by Intel. (You can find the agreement under which AMD is able to produce x86 CPUs on Google if you're really interested)

Of course, axel, lex, and giant are all really happy they were right about the performance of the processor, despite the fact that this means we may be losing Intels only motivation to innovate.

Happy? Not at all. I'd love AMD to release a new CPU that blows Intel's fastest away so that these low prices and the insane competition that is occurring now will continue. It just isn't happening unfortunately. If the rate of innovation and progress in the CPU market slows down I'm directly blaming AMD: They haven't raised the performance bar at all with the Phenom CPU.

Worst of all, according to the reports from Fudzilla (that were already quite well known) all AMD will have next year to fight Nehalem is a 45nm version of Phenom and Barcelona with a larger L3 cache.

As the many reviews show, Phenom has a lower IPC than Kentsfield. They need a higher clockspeed to take the performance crown. But that isn't happening.

I'll be brutally honest:- By the time Nehalem ships in Q4 next year, I don't think that AMD will have a processor that is significantly (10% or more) faster than Intel's original QX6700 quad core CPU from November 2006.

Ho Ho said...

giant
"Even if NVDA wanted to, the FTC wouldn't allow it, as it would give NVDA a monopoly in the discrete GPU market"

Well, Intel is also entering discrete GPU market soon


"Worst of all, according to the reports from Fudzilla (that were already quite well known) all AMD will have next year to fight Nehalem is a 45nm version of Phenom and Barcelona with a larger L3 cache. "

To be honest I'm not counting on Shanghai being availiable in quantities and decent speed before 2009.

Poke said...

So what happened? Phenom is obviously slower clock for clock compared to Intel's old Kentsfield. Phenom was pre-fragged by Kentsfield, pathetic.

Mo said...

So what happened? Phenom is obviously slower clock for clock compared to Intel's old Kentsfield. Phenom was pre-fragged by Kentsfield, pathetic.

If and When Scientia and Abinstein come out of hiding, they'll be able to tell us.

Aguia said...

So what happened? Phenom is obviously slower clock for clock compared to Intel's old Kentsfield. Phenom was pre-fragged by Kentsfield, pathetic.

Well that was already expected since it’s obvious despite being a dual core CPU, Intel processor is also designed to be the fastest single core CPU which is still the most important thing since most applications aren’t multi threaded and the ones that are, are still badly implemented.
For example the same cant said of the K8 dual core. K8 dual core = K8 single core in performance (single thread).

However I admit that I was expecting Phenom > Barcelona (Opteron)

Since the Phenom NB is still clocked at the same clock Opteron NB is maybe there is something to be fixed. Because where is the 3.6Ghz HT?
And the idea of going dual channel 128 bit, to dual channel 2 x 64 bit was good?
It was a great idea in dual core cpu where each core had its own IMC for example, but in quad I wonder?

gdp77 said...

phenoms should be priced 40-50% cheaper than Q6600, in order to have a fighting chance. The world+dog will just throw a C2Q in their 775 motherboard, so they need a strong motive in order to buy phenom + 7xx chipset.

Personally I don't see how AMD will survive the following months. Ati 3850 seems a nice product, but is it enough?

Pop Catalin Sever said...

"Well that was already expected since it’s obvious despite being a dual core CPU, Intel processor is also designed to be the fastest single core CPU which is still the most important thing since most applications aren’t multi threaded and the ones that are, are still badly implemented."

Oh really? So it's not K10's fault it's the dumb developers fault because they can't multi thread properly, is that what you are saying ?

Maybe those dumb developers that can't multi thread properly, should have optimized their applications for Barcelona long before the first chip made the appearance on the market. Well guess what, to optimize an application for particular micro architecture you need to have that chip, and actively develop on it. Starting from compiler writers to application writers. You need to do measurements, benchmarks and run tests for your particular problem domain.

Also most games that were build multi threaded in the past were build to use 2 processors/threads, guess why? because that was actually an optimization (using the precise parallelism the hardware has) instead of using an arbitrary number of threads that would have overly complicated the work division of the engines and thread synchronization.

But hey, we developers are stupid, we don't have a clue what we are doing, just because software doesn't magically run better on new hardware that appeared a month or two ago and which we didn't even got a chance to look at.

P.S. maybe you can enlighten us on how proper multi threading needs to be implemented.

Giant said...


Well, Intel is also entering discrete GPU market soon


That's true. But that's not until late 2008 at the earliest.

Personally I don't see how AMD will survive the following months. Ati 3850 seems a nice product, but is it enough?

The 3850 and 3870 are nice GPUs. Not quite as fast as the 8800 GTs, but they are cheaper to produce and are sold as such.

The thing is, you can run Crossfire on Intel boards. Many people will just avoid Phenom's mediocre performance and run Crossfire on an Intel P35 or X38 board. As such, I see boosted sales in the graphics division but that's not enough to offset massive losses coming from AMD's CPU business.

Ho Ho said...

aguia
"most applications aren’t multi threaded and the ones that are, are still badly implemented. "

Wasn't k10 with all its buzzwords supposed to be scaling way better than Core2 in multithreaded workloads, especially when they are badly implemented with too much traffic between cores? Come on, with Core2 cores have to sync their caches by going through northbridge but in K10 they can do it through L3. Still, real world confirms that Intel solution is at least as fast as AMD one, if not faster.

Aguia said...

Pop Catalin Sever,

Oh really? So it's not K10's fault it's the dumb developers fault because they can't multi thread properly, is that what you are saying ?

No they aren’t dumb they have just diverged.
With the introduction on the Xbox 360 and PS3 the developer’s efforts have been spent to improve multi core usage.
With the PC the 64 bit efforts had gone to the core usage too. With many computers already hitting the 2,3,4 GB has standard. Sooner or later they have to migrate to 64 bits applications. But that will take some years yet. So the priority is multi core not 64 bits.

In other words AMD 64 bit introduction was done at the right time however the very slow adoption of it, the consoles low memory quantity VS huge amount of cores, the orders are implement x core/thread efficiency. That is pretty obvious.


But hey, we developers are stupid, we don't have a clue what we are doing, just because software doesn't magically run better on new hardware that appeared a month or two ago and which we didn't even got a chance to look at.

If you are a developer and you say you are stupid I accept your word as granted. It was you who said you are stupid and developers are stupid not me.

Look Core 2 Duo was the fastest when it got out overall with new or old applications, so are you telling me they optimized 3Dmark 2001, 2003, and others programs even before the processor got out, amazing...
The processor (Core 2) is the fastest in single thread or multi thread applications and that is a fact, period!


Ho ho,

Wasn't k10 with all its buzzwords supposed to be scaling way better than Core2 in multithreaded workloads, especially when they are badly implemented with too much traffic between cores? Come on, with Core2 cores have to sync their caches by going through northbridge but in K10 they can do it through L3. Still, real world confirms that Intel solution is at least as fast as AMD one, if not faster.

Well maybe with the application that runs on servers might help. With desktop software it seems not. Let’s wait for some faster Opterons and we will see if it’s fast or not.

Ho Ho said...

aguia
"Sooner or later they have to migrate to 64 bits applications."

So, what kind of special care should developers take to make their 64bit programs work well? Only thing I can think of is using proper integer sizes and less pointers though that will probably help Intel more as all those two things do is use up more memory->generate more traffic and difference there is neglible.

Btw, I've been running 100% 64bit system for over a year now with all the programs besides Flash9 compiled as 64bit. I could have done it at least 1-2 years ago but didn't bother.

Btw, aren't both XB and PS3 running 64bit systems?


"Look Core 2 Duo was the fastest when it got out overall with new or old applications, so are you telling me they optimized 3Dmark 2001, 2003, and others programs even before the processor got out, amazing..."

Don't you think it just could be that Core2 was optimized to run average applications good? That also includes older ones. It is not like someone optimizes a CPU to a certain application or something.


"Well maybe with the application that runs on servers might help"

Have you seen any benchmarks confirming that? I'd say only bandwidth intensive stuff benefits from those buzzwords.


"Let’s wait for some faster Opterons and we will see if it’s fast or not."

Faster clock speed doesn't help with IPC. Only thing that could happen is that, yet again, thanks to higher memory bandwidth K10 scales better. Though majority of desktop apps are not bandwidth intensive.

Aguia said...

ho ho,

So, what kind of special care should developers take to make their 64bit programs work well?

Nothing special. I didn’t say it was too hard to do, do I?
I think you guys don’t read what I wrote and put words in my mouth that I didn’t say just because you want to discuss things that wasn’t said.
Now I understand how Scientia must fell sometimes...


Btw, aren't both XB and PS3 running 64bit systems?

Who cares, they won’t find a wall in the memory addressing (none of the consoles has more than 512MB of RAM), nor a change in Operating Systems and drivers like in the PC. PC and consoles are two completely different worlds. Besides I doubt they are using extensive 64bit instructions.


Don't you think it just could be that Core2 was optimized to run average applications good? That also includes older ones. It is not like someone optimizes a CPU to a certain application or something.

Did you hear this Pop Catalin Sever? It seems ho ho agree with me. That answer is for you.


Have you seen any benchmarks confirming that? I'd say only bandwidth intensive stuff benefits from those buzzwords.

Have you seem any quad Opteron clocked at 3.0Ghz?
If not then I haven’t seen any benchmark that confirms that.


Faster clock speed doesn't help with IPC.

Humm… So a Core 2 duo 2.0GHz runs about the same performance has one 3.0Ghz? Sorry but I didn’t understand.

Ho Ho said...

aguia
"Nothing special. I didn’t say it was too hard to do, do I? "

Exactly so I can't see what is so special in having developers write 64bit software. Basically the only thing they need to take care of is that they shouldn't assume sizeof(void*) == sizeof(int). There is nearly nothing about performance.


"Besides I doubt they are using extensive 64bit instructions."

Err, what kind of 64bit instructions are you talking about? Consoles are based on Power architecture. Instructions under 64bit x86 are pretty much the same as under 32bit.

"Have you seem any quad Opteron clocked at 3.0Ghz? "

No but I've seen them at 2.6GHz and there wasn't anything special about them. Why should I think that something special happens with some more added clock speed?


"Humm… So a Core 2 duo 2.0GHz runs about the same performance has one 3.0Ghz? Sorry but I didn’t understand."

What you do not seem to understand is what IPC is. My point was that K10 has lower IPC in majority of applications and it would need to be clocked higher to beat Core2 in them. There is no magic that could happen that would suddenly boost K10 IPC just because clock speed increased.

Pop Catalin Sever said...

"No they aren’t dumb they have just diverged."

Diverged from what? Please explain.

First you say: "since most applications aren’t multi threaded and the ones that are, are still badly implemented"

Then: "With the introduction on the Xbox 360 and PS3 the developer’s efforts have been spent to improve multi core usage."

So the developers made efforts to improve multi core usage (a.k.a divide the work as evenly as possible) but they still didn't made proper multi threaded applications.

Now if you please, tell me where the flaw in there implementations is. What's wrong with the multi threaded applications ?

"With the PC the 64 bit efforts had gone to the core usage too."

There always were efforts to improve the hardware utilization and optimize the software, this is not something new, and 64 bit didn't brought this. But this is a complex topic, you can't use SSE for example every piece of code there is because SSE is domain specific, meaning you can use it where you perform the same independent operation on multiple sets of data, and this is not the case with a large part of the code)

"If you are a developer and you say you are stupid I accept your word as granted. It was you who said you are stupid and developers are stupid not me."

Yes we are apparently... we can't multi thread properly to make K10 look good. Btw you still didn't expalained how the current multi threaded applications are badly implemented, if you do maybe we have a chance to redeem our selfs.

"Look Core 2 Duo was the fastest when it got out overall with new or old applications, so are you telling me they optimized 3Dmark 2001, 2003, and others programs even before the processor got out, amazing..."

You understood the exact opposite of what I said. I said:

"to optimize an application for particular micro architecture you need to have that chip, and actively develop on it. Starting from compiler writers to application writers"

and I don't have a clue where you pulled: "so are you telling me they optimized 3Dmark 2001, 2003, and others programs even before the processor got out"

Anyway you still need to explain how are are the multi threaded applications badly implemented.

Ho Ho said...

"Anyway you still need to explain how are are the multi threaded applications badly implemented."

I'm more interested to know how one can implement multithreaded software that works better on Core2 than on K10

Aguia said...

Anyway you still need to explain how are are the multi threaded applications badly implemented.

All is resumed to the scale problem.

For example how good do you think a good multi core program is?

Performance boost in %:

a-
1 core 100%
2 cores 120%
3 cores 130%
4 cores 135%

b-
1 core 100%
2 cores 140%
3 cores 160%
4 cores 170%

c-
1 core 100%
2 cores 180%
3 cores 210%
4 cores 250%

d-
1 core 100%
2 cores 170%
3 cores 250%
4 cores 320%

So let’s assume AMD K10 is better core efficient when all cores are loaded. Let’s assume developers are already in fase (b), Intel at that fase will keep a long gap between the AMD systems, but when they get to (d) AMD will remove the edge Intel got.

However if programs are still single thread and not yet multi core efficient (badly implemented maybe is outrage lets say not yet mature enough) who will keep winning?
I think when Intel and AMD starts using more cores in their CPUs and developers improve their applications to multi core we will see less interest in single/core thread performance.

Ho Ho said...

aguia
"So let’s assume AMD K10 is better core efficient when all cores are loaded."

What? You seem not to understand anything of what you were talking in that last post.


Also you seem not to know that there are apps out today that are "in phase D" and are still faster on Intel.

Pop Catalin Sever said...

Those phases you were talking about don't exist. The scaling of a multi threaded application is defined on how many threads it uses and how much of the time those threads do work. Meaning having a number of threads equal or greater than the number of cores and having those threads perfectly balanced (all threads doing work all the time) gives you the best scaling possible (nowhere else to go from there).

3DMax results:
QX6800 0:39
X6800 1:11
So dual core to quadcore scaling is 89% for a dual core that's at a lower frequency by 2.4% (That's phase e for 3D Max)

Unfortunately the same comparison can't be made for AMD because there are no K10 X2. But The Scaling from 2.4 GHz Athlon X2 64 (4600+) to a 2.4 GHz Quad core Phenom is 101% (1:47 -> 0:52) . Which doesn't mean anything because the cores are different and supposedly K10 has double the SSE power of K8 (note that 3DMax is highly SSE intensive). And with this high scaling Phenom still trails Intel's Q6600 at 2.4 GHz.

Benchmark data is from TH (Lastest Phenom article)

Mo said...

I don't think Scientia or Abinstein have gone this long without posting in this Blog.

This is shocking.

Mo said...

And about the QX9770 Thermals...
Well I hate to tell you guys (those who were hoping the chip would be HOT). Anand has an update and there is an issue with the board they were using. The board was not fully compatible with the chip and needed a bios update.


Updated: We're working with Intel on the source of the thermal problems we mentioned in this review, it looks like the culprit was our ASUS P5E3 Deluxe motherboard. ASUS has since released an updated BIOS intended to address the power consumption issues we faced, you can read more about it here while we continue with our testing.


http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3154

Wasn't it quiet obvious that There was an error somewhere? from the getgo?

David said...

Making predictions in the tech industry is like weather "forecasting". I look back on some of the older posts in this blog and I think the author had better switch to the weather report side of things (because who can go wrong with those?) and lay off on the weather "forecasts" for a while.

Mo said...

Making reasonable prediction is not that hard... making ones that are clearly out of reach are just ridiculous.

I remember it VERY well how Scientia blew me off when I said that we won't see 3ghz phenom parts until Q2 and he CONSTANTLY insisted that it takes 6 months from the time AMD previews something... which would have made it ohhh Early Q1...

All the data was OUT THERE. the Data that clearly showed that AMD is having trouble ramping but Sci refused to accept that data.

Well it is now very clear that we won't see any boost till late Q1 and MAYBE just MAYBE we'll get to 3ghz in Q2.

LIke I said, it's not hard to make a reasonable prediction.

Sci has also constantly stood ground on yields, he believes the yields are EXCELLENT...when They're anything but excellent.

Oh sci.

lex said...

Yields are broken at AMD

I call Yield a good die at the right speed. In this case Yield is broken by all measures

1) Total Die is lacking. How can they be fully converted but still no material new products shipping?

They are shipping minimum leading edge native quad-core. That tells me they haven't achieved the process stability to get an measurable volume of units with all 4 die functioning. Most funny as this is the company calling themselves true quad-core and making fun of INTEL's approach to put 2 dual cores togather a far easier task.

2) Horrible speed on Bacelona and Phenom. Lots of die at slow speed is bad enough, few die at slow speed is a disaster. you got a big multibillion dollar full of expensive new steppers and hundreds of million dollars of in process invetory and its has poor process yield and what little die you get out is slow.

AMD won't share yield numbers except fancy powerpoint foils from an ex. IBM executive. When has anyone from IBM known how to manufacture hundreds of millions of units. IBM doesn't make a thing.

AndyW35 said...

Looks like 2.4GHz Phenom is postponed to 2008 from December, check out Digitmes front page.



Comments on 45nm being delayed do not surprise me either. Although I am not the worlds biggest expert on these things, or even compared to other people on tbhis blog I admit, the SOI process has concerned me for a while yet. Or at least AMD's SOI process. It has not been terribly good results at 65nm on K8 and seems very leaky for K10, you can only assume 45nm will be at least as tricky if not more so with increased leakage or the trade off to reduce this with too much heat generated instead.

Although AMD's architecture looks fine Intels process looks a lot better. They have really nailed it with their change for 45nm it seems. Good decision on their part.

November 22, 2007 12:19 AM

Ho Ho said...

mo
"I remember it VERY well how Scientia blew me off when I said that we won't see 3ghz phenom parts until Q2 and he CONSTANTLY insisted that it takes 6 months from the time AMD previews something... which would have made it ohhh Early Q1..."

Hehe, I remember having a few discussions about the same subject with him. I said "3GHz in late Q2/H2, if lucky". Well, so far it seems my "forecasts" have been much more accurate than his, assuming AMD will keep to its schedule and releases 2.6GHz at the end of Q1.

Just for fun I say we won't see much more 45nm AMD cpus in 2008 than we saw 65nm K8's in 2006. We shall see how accurate this prediction is.

Aguia said...

Hehe, I remember having a few discussions about the same subject with him. I said "3GHz in late Q2/H2, if lucky".

Let’s change that a bit.
How about AMD got unlock with that TBL problem and can’t release anything faster.

If AMD didn’t have that problem you, mo and lex where eating your own words.

PS: You all talk like Intel never had happen such a thing, when it happened a lot of times and it took them months to years to solve them.
I know ho ho wants proof of what I said, he never remember anything that had run wrong at Intel. You can all find it here:
Http://www.google.com

Ho Ho said...

aguia
"How about AMD got unlock with that TBL problem and can’t release anything faster."

Didn't that problem occur with those "non-cherrypicked" 3GHz quads they showed months ago? If not then they must have added that problem with some later steppings. Also fact is that AMD promised to fix it by early Q1 when they are launching their 2.4GHz quad. As that problem will be gone by then why are they launching only 2.6GHz at massive 140W TDP late Q1? By then there is no TBL errata to blame for their clocking and heat problems.

Also I know Intel has had problems, everyone has. It is not like AMD is somehow special with theirs. It is just that they fail to admit that they are having huge issues with K10 and likely with 65nm in general. It is quite similar to Intel and Prescott.


One fun thing to do is to plot K10 clock increase in time. Let's see:
10'th September: 2.0GHz
15'th November: 2.3GHz
Early Q1: 2.4GHz
Late Q1: 2.6GHz

Basically it seems it takes them around one quarter to increase clock speed by 300MHz. Of course things seem to get tougher and tougher as time goes by and 125W for 2.4 and 140W for 2.6GHz are proofs of that. If they by some miracle are able to keep that 300MHz per quarter speed increase then they should get 3GHz by around mid-Q3. Of course if their thermals also keep increasing as fast that 3Ghz would be somewhere around 180W.

Of course this is just some very simple prediction made from a few datapoints and has no value besides enterntainment. I'm sure things won't be exactly as I describe. I doubt AMD would release a CPU with >140W TDP and I have no ideas if they can increase their clock speed as fast. Only thing one could conclude from that timetable is that it will be awfully difficult to release 3GHz quad with decent thermals in H1 next year.

Aguia said...

ho ho,
I agree with you.

Tell me one thing where did you get the 140W figure?

Do you base that on this?
http://www.legitreviews.com/article/597/2/
(last slide)

«Oldest ‹Older   1 – 200 of 218   Newer› Newest»