Tuesday, December 12, 2006

Walking Into A Mirror

The most significant development this year in the competition between Intel and AMD seems to be that the days of moving in different directions are over. K6 was not so different from PIII. However, when AMD moved to K7 and Intel moved to P4, the architectures diverged dramatically. As AMD continued with K8 and Intel moved to Prescott, the architectures became even more different. By 2005, in terms of both architecture and corporate structure, there was little overlap between Intel and AMD. This year, with the introduction of Core 2 Duo Intel has pulled its architecture back in the direction of K8. Likewise, AMD is moving in Intel's direction. What we will see through 2007 and 2008 will be the convergence of the two companies both architecturally and structurally.

In 2005, AMD and Intel were as different as they had ever been since AMD started making X86 processors. AMD had only a single 200mm FAB compared to Intel's many 300mm FABs. And, having divested of Spansion, AMD was strictly processors with a few server chipsets in contrast to Intel's diverse interests including its large chipset and embedded graphics business. In terms of cpu architecture, Intel had a long pipeline, virtual HyperThreading, high clock with low IPC design. AMD's short pipeline, dual core, low clock with high IPC design was almost completely the opposite. Intel's MCM Smithfield procesors were a small step in AMD's direction. However, in 2006 with C2D Intel moved to a short pipeline, high IPC, true dual core design and dropped HyperThreading. The architectures are still different but not nearly as different as they were last year and they are still converging.

Just as Intel moved closer to AMD by shortening the pipeline and increasing IPC while extending the architecture to 64 bits so too AMD nows moves in Intel's direction. Core 2 Duo has a dedicated stack unit and AMD moves closer to this with its sideband stack optimizer. AMD improves instruction reordering and out of order loads moving it closer to Intel. AMD now has data dependent latency on the divide instruction as Intel had already done. AMD stretches SSE width to a full 128 bits doubling the speed of SSE operations. It doubles the bandwidth of Prefetch and the L1 bus. Again, these are changes that Intel had made from Yonah (Core Duo) to C2D. AMD might have a small advantage here due to twin L1 buses rather than just a wider L1 bus.

Intel's architecture has been better with DDR2 because of the larger cache. AMD is now moving to a split memory controller where each half can act independently which is what Intel has now with two memory channels. AMD also increases cache size by adding shared L3.This along with improved memory scheduling, a better write burst mode, and out of order loads more than makes up for the higher latency of DDR2. We should see a big improvement for AMD with DDR2 and into DDR3. This again, moves AMD up to where Intel is now.

AMD also moves in Intel's direction by adding some inclusive functionality to its cache while also adding shared cache as Intel has now with C2D. AMD is expanding the number of FastPath instructions. This is important because the decoders can do 3 FastPath instructions or 1 complex instruction per cycle. The more instructions that are fastpath the better. This again makes up for some of the improvement that Intel gained by adding a fourth decoder and macro-ops fusion.

Overall, AMD moves much closer to Intel's C2D with the changes in K8L. AMD will add to its already greater memory access speed and match or beat C2D in terms of SSE. However, the changes don't quite add up to C2D's 4 instruction issue and Macro-ops fusion. I would guess that AMD will remove about 3/4 of the current IPC gap. This would mean that a K8L of 3.1 Ghz would match a C2D of 3.0 Ghz. This is much better than today where it would take a 3.5 Ghz K8 to match the same 3.0 Ghz C2D.

Both AMD and Intel use independent L1. AMD's independent L2 plus shared L3 should be pretty close to Intel's large shared L2. The main differences today are that Intel's C2D still does not allow I/O access above 32 bits and I'm not certain that C2D supports addressing beyond 36 bits. AMD has the advantage with an onboard memory controller and point to point interface. It looks though like Intel will defintely move in this direction with its own onboard memory controller and CSI point to point interface on a new socket in early 2009. AMD has also given indications that it is trying to shorten Intel's smaller process lead.

Structurally, AMD has taken big steps in Intel's direction by creating a 300mm FAB and purchasing ATI. ATI gives AMD a directly competitive position with Intel in terms of chipsets and embedded graphics. AMD will now be able to satisfy the needs of customers who want a complete solution from one manufacture as Intel has provided for some time. AMD will now compete with Intel directly in both embedded graphics for notebooks and embedded graphics for corporate customers. At the same time, Intel has begun divesting of underperforming sections of the company and trimming down toward the main cpu business closer to where AMD already is. AMD's single 300mm FAB greatly boosts its production capacity while reducing cost. Upgrading FAB 30 to 300mm will continue to move AMD in Intel's direction.

There are probably AMD fans who are hoping that K8L will allow AMD to leap ahead of Intel's C2D in 2007. Similarly, I imagine there are Intel fans who are hoping that Intel's move to 45nm will allow Intel to stay ahead of AMD. However, neither of these scenarios seems likely. K8L should be pretty even with C2D architecturally rather than a leap ahead. For process technology the rumors abound with stories of 4.0Ghz clock speeds for Intel while decreasing power draw. These same rumors often limit AMD to as little as 2.9 Ghz putting AMD one full Ghz behind Intel. This would put Intel significantly ahead of AMD if true but it doesn't seem likely. AMD shares process technology with IBM. It is certainly in IBM's interest to be competitive in terms of process. So, it seems likely that if something were lacking that would allow Intel to gain significant advantage IBM would find another company with additional process technology that could be licensed. I'm reminded of the many rumors that Intel would have 3.2Ghz C2D chips at launch but these are still unreleased. I'm also reminded that after Revision F of K8 was released there were claims that it was actually slower than Revision E. It seems that optimism tends to follow Intel while pessimism follows AMD regardless of what each company actually does.

I expect AMD to continue to slowly gain share at the rate of about 0.66% per quarter. I expect AMD and Intel to be roughly equal on the desktop in Q3 07. Likely Intel will have to address AMD's new dual socket FX once AMD has quad core chips for these. If Intel did not address this area they would have nothing competitive. It also looks like Intel will finally have a 4-way server C2D offering in the same timeframe. However, just as AMD's quad FX won't reach its full potential until AMD has quad core FX's, Intel's 4-way server chipset won't really pull Intel even with AMD. Clearly, this will be a stopgap until Intel can release a new socket with integrated memory controller in 2009. However, Intel has no choice but to offer a stopgap chipset since the P4 Xeon will be completely outclassed by K8L. Intel's processor will be fine; it will be the chipset that is stopgap. Intel's quad FSB northbridge will be slower and more expensive than the equivalent Opteron 4-way but it should be close enough to stop Intel's 4-way server share loss. Intel will probably get a nice boost with 45nm technology that will give it an edge on the desktop late in 2007 but AMD will follow with its own 45nm technology by mid 2008. However, any advantage that Intel might have will tend to be countered by AMD's increased competition with embedded graphics. But this won't be the only change. Increased revenue gives AMD more money for R&D and AMD's advantage with APM will come into its own once AMD has two 300mm FABs. My guess is that Intel will also be taking steps to try to match APM and will probably have to do a second reorganization in 2008. AMD for its part may begin construction of a third FAB, possibly after mid 2007.

We know that Intel will move to an Integrated Memory Controller moving it towards AMD. It would seem reasonable that AMD must be looking into Intel's 4 instruction issue and macro-ops fusion to gain future speed. We also know that Intel will eventually use SOI as AMD has been doing for years now. By 2009 AMD and Intel should look much more similar than they did in 2005. I imagine the two will continue to converge as well go into 2010. For example, Intel is moving to a 2 year core design cycle while AMD is moving to a modular core. Intel's approach is less efficient than AMD's however Intel has more engineers to throw at the problem so I would imagine the results would roughly match. If 4 instruction issue really is a good idea then I would expect AMD to add this perhaps in late 2009. There will be a slight difference in process with Intel using high K dielectrics. I don't see any problem with this. Intel only needs to find a high K material with a suitable bandgap for this to work. The only area that I'm not so sure about is Intel's tri-Gate design. From what each company has said it looks like the tri-Gate will not be able to be shrunk while AMD's finFet will be. Still, one would assume that if the tri-Gate can't be shrunk that Intel would follow AMD and use a 3D dual gate like AMD's finFet. If AMD does build a third FAB the X86 processor market should turn into a true duopoly with much more stable prices and production. The big question is whether the market will reach duopoly status with just two 300mm FABs at AMD. However, it seems a certainty that into the foreseeable future AMD and Intel will continue to converge.

35 comments:

Scientia from AMDZone said...

This is a document on Intel's website in the Technology and Research, Silicon section:

Staying Ahead of the Power Curve

So does Intel rule out SOI completely?

No. There is a type of SOI called fully depleted (FD-SOI) that has merits beyond the partially depleted SOI that some
of our competitors are using today. FD-SOI has been under active evaluation for some time at Intel.

Scientia from AMDZone said...

I am still baffled why anytime I show scepticism for rumors about Intel technology I am immediately labeled an AMD fan. I am also baffled by the suggestion that I actively seek bad news about Intel and good news about AMD.

In reality, both Sharikou and Sharikou180 post cherry picked news items on their blogs. Sharikou does indeed post items that are bad for Intel and good for AMD while Sharikou180 does the opposite.

If all I wanted to do was show unrestrained enthusiasm for AMD, I could certainly do the same thing. But I don't. I have never posted cherry picked news items to try to create a positive spin for AMD. I try to take the totality of what is going on rather than just an item here or there.

It could very well be the case that Intel is holding back on 3.2Ghz chips simply because AMD is not challenging them enough right now. It could also be the case that Intel's 45nm process technology will give them a significant boost in speed an power efficiency over AMD.

To clarify Intel's current position: Intel is clearly ahead by 15% IPC on integer calculations and ahead by much more with SSE. AMD is ahead with FP calculations but these are clearly becoming less and less important. AMD is probably competitive with non-SSE memory intensive operations but these have not been adequately tested.

Intel's advantage is greatest on the high end desktop range. I don't see AMD's FX chips including quad FX as being directly competitive with X6800 and Kentsfield. If AMD did not have plans to offer a much more power efficient and more powerful K8L quad core FX chip then I would say that 4X4 was a mistake. For much of this we will simply have to wait to see what rumors are true. We should know by the third and fourth quarters of 2007.

ashenman said...

Another good article.

Question (or at least a feeling for the need of clarification), what exactly would make AMD's APM system more efficient than Intel's current system? With fewer fabs APM makes sense because you can extremely efficiently control product output. But with more, you become limited by the output of the individual fabs, while copy exact allows you to increase focus in more than one fab on a single product. Maybe this is just because Intel holds such a significant portion of the market that they find a need to do this, but it would seem like from the stand point of mass flexibility, Intel's scheme is somewhat better.

On the point of bad news always following AMD, this is just Intel's brand recognition kicking in. It's human nature to be skeptical of something you are uncomfortable with because of a lack of brand recognition. AMD will thus always have worse press than Intel, because investors and people who do not follow the market and selectively (instead of intuitively) choose which product they use will have immediate bias towards Intel.

Scientia from AMDZone said...

No. You are misunderstanding about APM. APM is more sophisticated than what Intel is using. Intel has some process automation but most of it is simply involved with order and scheduling and not the actual manufacturing process itself.

AMD's APM watches everything at each tool in the process. APM is capable of adjusting the process and correcting errors. This is more flexible than what Intel does rather than less fexible. Also, APM becomes even better when AMD has two similar FABs rather than one 200mm 90nm FAB and one 300mm 65nm FAB. If AMD builds a third FAB in NY it will become immediately available in terms of production control. Intel does not have this same level of sophistication. However, Intel could be working on it.

gdp77 said...

If AMD did not have plans to offer a much more power efficient and more powerful K8L quad core FX chip then I would say that 4X4 was a mistake.

Imo 4x4 is pointless atm and made AMD's name a synonym for "joke", all over the web. However, 4x4 can be transformed into an overkill platform when K8L is released. 2GB dimms will be available then, so 8 cores + 16GB of RAM + NUMA enabled OS (Vista 64 bit) will give to AMD the performance crown. Intel will have no chance to respond to this kind of computing power in desktop, until 2009 (unless I am missing something).

ashenman said...

Red, is your only comment concerning this specific article that it's too long? You really need to stop complaining about that as all it shows are your own personality defects or a case of ADD. I don't know why scientia didn't just add those two posts into his article, but then again, maybe he felt it made more sense for them to be in here.

Whatever options AMD has for quad fx have already been considered and decided upon. It's not like they didn't know how their own platform would perform or how much juice it would use months in advance. Their pricing changes reflect this as they changed their pricing on the 4x4 lineup technically before there were any reviews. This shows they kept the price up despite kentsfield in order to inflate demand from distributors.

I don't see how he's "hyping" AMD, and I don't see how he's "downplaying" Intel. Most people see the extreme results from this competition. For some reason, some people see that AMD is going to be pushed back to where it was, and even though he says he doesn't believe that, 180 basically says AMD's growing trend will reverse, which will only be able to put it back where it started off because of demand and loss inflation due to the highly unstable nature of this market. Sharikou says Intel will end up filing bankruptcy. Scientia obviously sees the middle ground with the situation becoming what is essentially a healthy duopoly by 2009.

Red, why don't you calm down for a while, and maybe just ignore this blog for a week if you can't say anything constructive.

Anonymous said...

Red, think of blogs like people's houses not like public squares.

Every blog owner has the right to delete or disallow a any comment whether fair or not from a objective point of view.

It's of no use complaining that posts are deleted. Instead I suggest reconsidering why are they deleted ...

Anonymous said...

I've said many times that his articles are too long. He got mad once but then his own little groupie came and basically said what I said, that he had no idea what he was saying. I think one does not need to have 50 details to say that Intel and AMD are becoming more alike. Perhaps you should step into the shoes of someone that randomly found this blog. If faced with this wall of text, like others, I would probably be very confused and lack the enthusiasm to digest it all.

Back to name calling? I know those exams are tough, but even if I did have ADD, what have you accomplished by accusing me of it? And what would you make of it if I did? :/

..Quad FX is still going to be extreme wattage, are they going to sell 8 cores for $1000 max, where are the apps and where is the need?

I think Intel will have continue their reversal until mid07. It is anyone's game from that point. Complaining about Intel 'paper launches' but not acknowledging AMD's recent ones, complaining about THG but not AMDZone, telling us what a great gem Transmeta would've been many years as it supposedly shows how silly Intel is in the face of actual news, blah blah. You are right, you are totally impartial. I totally don't feel bothered to argue about this, but Scientia seems deeply hurt that he is not viewed as not biased.

Pop, you said it yourself, whether fair or not, blog owners can delete what they want. That doesn't make it right.

Erlindo said...

As always, great post Scientia. I do enjoy reading your blog a lot.

I'm a bit puzzled with the following:

When would we see a "new" core from AMD (just the way they did with K7) that will put intel back to the drawing board?

I'm asking this becasue you seem kinda lukewarm that AMD's upcoming Barcelona will not outperform intel's C2D architechture, but instead, it will match it and "exceed it" in some special scenarios.

I'd hate to know that AMD will be copying intel's architectural approaches. Wouldn't be better for AMD to make their upcoming "next-gen" processor an 6-8 issue instead of 4 like intel's C2D?
Would it be impossible to achieve this in a certain way?

Thanks in advance for your time.

P.S: Don't listen to the trolls

Anonymous said...

This article is convoluted..? Nicer? ;) And yeah, you might want to tie the 1st comment into your article. And 2nd comment? o.o


http://www.theinquirer.net/default.aspx?article=27421
They have tried already my constantly personally jabbing not friend :D

But if you insist on changing the discussion this early.. 180 hypes Intel more than you hype AMD. Better to be optimistic than to be a downer. And talks about the present, oh, about..99% of the time:) 180 posts stuff that many others have reported about, hardly cherry picked. Maybe it's just me, but this blog often digs deep into the past, future, or into the most obscurest anti Intel trenches. So please explain why you have extremely long posts dedicated to downplaying Intel and none for AMD yet claim to be impartial.

http://www.pcpop.com/doc/0/168/168366.shtml
http://www.xtremesystems.org/forums/showpost.php?p=1888383&postcount=159
If comment 2 is allowed..:D

ashenman said...

Very good job red!! (I just got back from my last final if you can't immediately tell). That's a much better way of approaching your dislike for the length of scientia's articles, and is a fault I also share (you've probably noticed).

That article is just charlie being charlie, and while I'm not saying it's impossible for him to be right, he shows no real proof except for the fact that there's no news on it, which is like writing a news article about the lack of news lately (sound a little pointless?).

Talking about the past and relating it to the present in very complex and convoluted arguments may seem terrible, but talking about the present most of the time without respect to the past or with only passing mention to broad assumptions that are not necessarily accurate about the past (quite a bit of, but not all of what 180 does) is definitely no better.

I don't see how his posts are all about downplaying Intel. All he's done is compare both companies pasts.

I actually don't think netburst was a mistake, now that I look back at both companies histories. Netburst was simply a crutch that Intel just had a hard time getting off of. When AMD started competing effectively with Intel on clockspeed and process size, Intel had to do something to regain the clockspeed crown, and netburst's amazing error tolerance allowed substantial clockspeed increases that were easy to market. Which brings us to where we are now.

Scientia from AMDZone said...

Okay, this is the situation. The first 4 comments included one that said that Intel was not moving to SOI. Two of the comments were essentially a personal attack saying I was just an AMD fan looking at the world through distorted green colored glasses.

So, I deleted the comments that concerned my biased view. I would tolerate this more further down in the comments but not at the very top. In other words, the comments will not open with personal attacks. However, I didn't want to dismiss what had been said so I did two comments: one covering SOI from Intel's website and another explaining why I don't feel that my blog exists soley to pump AMD and shoot down Intel.

However, the original posters apparently were insulted at having their attacks deleted and they deleted the relevant comments as well. Then of course, Red came along and made it worse by doing a long comment about why I had posted the first two comments.

I'm not trying to be a blog tyrant but I would prefer that someone who says I'm biased give a reason. Am I missing something or distorting something? I get the impression that unless I put a neon banner on the first page that says, "Intel Rocks!" someone is always going to claim that I am biased. The only thing I can think of is that I haven't yet posted an article that is really upbeat about Intel. That might be possible if I can give it enough depth.

Anonymous said...

The problem is the quality of the posters is sorely lacking. I always like Scientia's blog posts as they are often very nice and balanced and informative.

The problem is that people do not seem to have the relevant technical knowledge and when corrected, they seem to resort to ad hominen arguments to cover up their incompetence.

Azary Omega said...

Keep writing them long once. And make sure you do it at least once a day. Otherwise i wont have anything to read when i get home from work.

Scientia from AMDZone said...

I've said many times that his articles are too long.

You are correct, most blogs are 1-3 paragraphs per post rather than 1,000 - 2,000 words. I'm just not sure how to make them smaller.

but then his own little groupie came and basically said what I said, that he had no idea what he was saying.

I'm not quite sure what this means. However, if I don't know what I'm saying why do you read my blog? I hope your are incorrect because I've had 8,500 different people look at my blog in the last 3 months so I'd be misleading a lot of people.

..Quad FX is still going to be extreme wattage

Very true.

, are they going to sell 8 cores for $1000 max

I would say, yes. I would say that Intel will too.

where are the apps and where is the need?

Well, maybe MS could devote one core to cleaning up discarded resources and keep their OS from bogging down; this happens all the time with XP. If you want the straight the demand for more cores is there; the reason you don't see it is because MS no longer has the flexibility to do proper research. Back in the early days of Windows they had two direct competitors and had to be more on their toes. Someone like Digital Research would have the answer if they hadn't been gutted by MS. It will take longer but it will happen.

I think Intel will have continue their reversal until mid07.

What is Intel reversing? They aren't reversing volume share. If you are talking about microprocessor revenue then it will probably take Intel till the end of 2007 to make it back up to where they were in 2005.

complaining about THG but not AMDZone

AMDZone doesn't do reviews. However, they could probably start if you can convince Intel to send them preview hardware.

supposedly shows how silly Intel

This is a good example of how you misunderstand things that I write. What I said in "Roads Not Taken" suggests that Intel missed an opportunity; this doesn't make them silly.

Scientia from AMDZone said...

Erlindo

When would we see a "new" core from AMD (just the way they did with K7) that will put intel back to the drawing board?

Well, the next thing is obviously K8L. AMD was never far ahead with K7 against PIII and Williamette. The problem now is that K8 is more than one speed grade behind C2D. After that it changes, AMD will make incremental changes inside the modular core.

I'm asking this becasue you seem kinda lukewarm that AMD's upcoming Barcelona will not outperform intel's C2D architechture

Well, my best judgment is that AMD's approach is not enough to match Intel in integer but should bring K8L a lot closer. This is based on what I've heard about the architecture. I still think AMD could pull ahead in SSE although others seem to think that Intel will still still be faster.

I'd hate to know that AMD will be copying intel's architectural approaches.

There is no doubt that both architectures are becoming more similar.

Wouldn't be better for AMD to make their upcoming "next-gen" processor an 6-8 issue instead of 4 like intel's C2D?

No. The gain from 4 issue is small; the gain from 5 would be even smaller. The gain from 6 or more would essentially be zero without other fundamental changes.

Both AMD and Intel are going to hit the wall soon, probably by 2009. So, it should be interesting to see how each tries to maintain performance improvements.

The wall comes from being at an end on things. Bits, they won't go to 128 bits integer for example and SSE won't go to 192 or 256. Cores, they will stop either at four cores or maybe eight cores but 16 cores is not likely. This would mean that increases in speed would come from clock increases only unless some novel changes are introduced.

Ho Ho said...

"I expect AMD and Intel to be roughly equal on the desktop in Q3 07. Likely Intel will have to address AMD's new dual socket FX once AMD has quad core chips for these. "

You could say that MacPros are the answer. Though technically they are just relabelled 2P Xeons. Good thing about them is you can use quadcores in them today, only you can't yet buy them with quadcores.

"I still think AMD could pull ahead in SSE although others seem to think that Intel will still still be faster."

What exactly makes you think they might be faster? Is it something to do with cache bandwidh or something else? AFAIK, they should both have equal SSE throughput at same clocks, only difference could be when one of them can't deliver enough data to crunch.

Anonymous said...

I thing future CPU's should have 4 x86 general purpose cores and 12 geometric and stream accelerating cores.

Games can benefit from geometry accelerators in areas like physics, graphics even sound effects ...

Speech recognition and synthesis should arrive. Our word processors have stayed the same for too much time. There's nothing to improve unless you can make them understand speech.

Image and shape recognitions is still way out of reach because of processing power. Another appliance for a stream processor.

Computers of tomorrow definitely need eyes and ears and mouths otherwise they have almost all of processing power anyone could ever want.

AMD definitely seems to go in the right direction by integrating specialized cores on the same die with general purpose processor cores.

Steel Smack said...

I love the way Scientia claims Intel's C2D will still beat AMD's new K8L and you pathetic Intel fanboy's have the audacity to call Scientia a fanboy. Come on. He uses logic and intelligence to determine what direction he thinks the industry is headed. I suppose if you are a fan of either Sharikou or Sharikou 180 than logic and intelligence are not something you possess.

My opinion is that K8L will beat C2D clock for clock, but it's good to see someone who has an objectionable view as opposed to you guys who think Intel is perfect, and AMD can't possibly *almost* catch up to Intel in the next year.

Remember, it took 5 years for Intel to finally catch up to AMD. Intel still has to prove that they are top dog in anything but market share. If AMD passes Intel up next year and holds it for another 5 years (or 3, basically any substantial amount of time), it will be viewed as if C2D was a fluke, and Intel is truly not the company you all wish it was.

Erlindo said...

Scientia wrote: No. The gain from 4 issue is small; the gain from 5 would be even smaller. The gain from 6 or more would essentially be zero without other fundamental changes.

============================================

My understanding about this was that an n-issue processor can process more instructions per clock cycle (please correct me if I'm wrong).

Anonymous said...

Steel Smack said...

If AMD passes Intel up next year and holds it for another 5 years (or 3, basically any substantial amount of time), it will be viewed as if C2D was a fluke, and Intel is truly not the company you all wish it was.

Good point, but it does lead me to a few questions, and I hope someone has some clear answers.

* Note *
I am only 2 years into the technology (MB, CPU, GPU) and I don't have any knowledge prior to Netburst or K8.

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

1. Prior to Netburst and K8, who had the better processors?

2. Most people have said is that Core 2 Duo is a suped up P3, if that is the case, where would Intel be today if they had not gone the way of Netburst?

Please try and give some theoretical answers if possible on question #2.

Thanks.

gdp77 said...

are they going to sell 8 cores for $1000 max

I would say, yes. I would say that Intel will too.


Are you saying that intel will somehow release a 8 core cpu @ 45nm? How will Intel be able to do that using FSB architecture? The only way I see that being possible is by releasing a dual FSB chipset. Do you have some kinf od information we don't ?

TheKhalif said...

Both AMD and Intel are going to hit the wall soon, probably by 2009. So, it should be interesting to see how each tries to maintain performance improvements.

The wall comes from being at an end on things. Bits, they won't go to 128 bits integer for example and SSE won't go to 192 or 256. Cores, they will stop either at four cores or maybe eight cores but 16 cores is not likely. This would mean that increases in speed would come from clock increases only unless some novel changes are introduced.



Excellent point. AMD will manage to get much closer to the theoretical 3IPC max which is better than redesigning for 4 IPC and having to improve effiency on four decoders.

because they use 3 complex decoders, they can optimize for that with branch history improvements and the like but by 45nm, the only choice foreither company will be HW threading.

I have predicted before that the arch after Barcleona would be 2 threads per core. At 45nm it would be doable and would mean less rearchitecting.

By using an arbiter with common logic, they can add 50% die space and get at least 80% increase. they talked abotu this previously and other than the "mysterious" RHT, that will be the only way without extending decoders.

I also saw a Stanford (IIRC) paper that stated that placing a FIFO pre-scheduler in the path you could get a 3 decoder arch up to nearly 6 IPC. I'll see if I can find it.

But mark my words HW threading is the next step for X64.

ashenman said...

Enumae, if Intel had not gone to netburst we would have probably seen a lot of mirroring much earlier, and Intel and AMD would have about equal marketshare right now. AMD had the ability to get much higher clockspeeds out of their architecture on .18um, which is what lead intel to develop netburst, in order to allow much greater error and heat tolerance with a very stable and very fast architecture. To do this, they had to use longer pipelines, as you know. At p3, Intel and AMD were pretty much neck and neck. Before that, AMD was normally behind, but always cheaper. Wikipedia actually has some pretty good articles on it, though both Intel's and AMD's need to be rewritten to avoid bias (though obviously people are jumping all over the AMD one).

Anonymous said...

Maybe off topic, and maybe some of you have read this, but Bloomfield is (possibly) Quad Core & Octo Threads, while having an IMC and CSI.

Octo Thread's

and...

IMC and CSI

ashenman said...

By the way, everyone who is bringing up the specialized cores is dead on, and that's what pretty much every processing company is trying to do right now. Either become a specialized core that can run next to an opteron or xeon, or get specialized features in extra cores.

Anonymous said...

Thanks Ashenman.

I will check out Wiki.

Anonymous said...

Try putting yourself into the reader's shoes?

He(reader) did not understand he(Scientia) was what I meant.

If Newegg can get away with selling QX6700s at those exorbitant prices, Nvidia with the 8800GTX.. Surely AMD can if their quads are going to be so superior..But if they feel that they have to compete by pricing their new fastest quad at worser margins ($125 per core) than their previous slowest ($150)..

So basically what you're saying is that there will be no apps to take advantage of 8 cores. :D So why would this platform with little (already hard enough for quad) support be competitive with one that performs as needed in the majority of apps but runs significantly cooler?

http://news.com.com/Intel+gains+server+share,+AMD+gets+notebook+boost/2100-1006_3-6130795.html
http://www.vnunet.com/vnunet/news/2163452/intel-holds-marginal-lead-amd
Retail, desktop, server..?

Actually, AMDZone does. Try looking on the site;) PlanetX64 seems to have no problems getting Intel review kits. Too bad for them though, according to a mod, AMD didn't like their positive Intel reviews so no Quad FX for them. Again, out of dozens of review sites, you continue to nitpick on a certain few.

And you also make out Transmeta to be a gem. You have the time to write all of these articles dedicated to Intel's 'blunders', but can't think of any companies AMD could've swallowed?

Wow, that AMD Wiki is trash. And some wonder where all these 'supposed' fanbois are xD

Anonymous said...

"But mark my words HW threading is the next step for X64."

Why do you think HW threads are such a big deal?

The next big deal is the fusion of x64 instruction set with that of ATI's and the standardization of this new instruction superset.

That will be the first step towards specialization of GP CPU for accelerating geometric computations way beyond SSE# and bringing massive parallel processing resources into every application's reach. Let's just hope compilers and programing languages won't be left behind ...
As I said before the software of tomorrow will need allot more streaming/geometric processing power than x86-x64.

x86-x64 only let's you add multiply subtract and move data around in very small bits. Streaming SIMD extensions allow you to do this on more pieces of data at once but that's just about it... CPU's are sequential (...I know about out of order bla bla) and all parts of them are dependent, all work on the same stream of instructions. This is the biggest bottleneck in x86-x64 and this needs to be changed quickly before processors run into a performance brick wall when the number of cores exceeds a certain limit... I mean maybe it's a good idea to make a car that can be driven by two drivers splitting tasks among them but is it an equally better idea to make a car that is driven by 100 drivers?

ashenman said...

catalin, I think that's what scientia was referring to before, and that's why specialized cores are such a big deal.

Red, AMDzone hasn't ever done a CPU review, they only link those. Other products, yes, but not CPUs.

I can't tell what your second paragraph is trying to say, please rephrase it.

Why would dual core have any support right now? Almost no apps support it right now save for quake 4 (which is still a very poor threading job). It's "future proofing" which in the computer business is very marketable, regardless of how intangible an item it is, and how much of a contradiction it is in such a marketplace.

Red, if you're so mad about how he nitpicks a few review sites, then please post as many other review sites that you think are good as possible whenever in any way relevant? It would be a very constructive activity, and all of us here would greatly appreciate it.

AMD has never had enough capital to merge or buy any other companies until they bought ATI, which was a very good choice, so their missteps would be purely in terms of their product line. These are all very well documented missteps, and going over them again would be mostly pointless since much of AMD has changed since then.

As such, there's never been any point where going referring to them would have been worthwhile as far as I can tell, though if you think there has been, why don't you start your own blog and write about it, so we can discuss it over there. Or just comment here about where you think it should've been done.

Anonymous said...

http://amdzone.com/modules.php?op=modload&name=Sections&file=index&req=listarticles&secid=6
..?

The Anandtech entry sounded rather long winded to me. And about Toms, seems suspicious that they don't OC AMD CPUs as much as Intel's but it's not like they're fudging numbers and then again, they don't even have E6300 numbers. Lastly, there are dozens of review sites. I don't see the reasoning for the vendetta against 2.

I think SLI and dual socket 'enthusiast' is silly. They should push 8 cores when they can fit it on one chip. And by the time that comes, apps would've moved onto quad? and so forth. Another thing for Scientia, with dual slots, GPU prices have not stayed the same over the years but have increased.


I'm still confused on the reasoning behind buying ATI.. I know they need a platform, but $5.4B? Keep saying that Transmeta would've been a gem:)

No comprende on the last paragraph.

TheKhalif said...

The next big deal is the fusion of x64 instruction set with that of ATI's and the standardization of this new instruction superset.

That will be the first step towards specialization of GP CPU for accelerating geometric computations way beyond SSE# and bringing massive parallel processing resources into every application's reach. Let's just hope compilers and programing languages won't be left behind ...
As I said before the software of tomorrow will need allot more streaming/geometric processing power than x86-x64.
Someone already posted a link to Intel going to 2 threads per core and I think AMD will follow suit. Most applications don't directly use fp matrices for calculations.

SIMD only works for certain things. Sometimes you just need two independent threads working together.

AMD presented an SMT framework back in June for their 2008 architecture. No real details have come out but it sounded like it will be HW threaded.

We'll see though.

Scientia from AMDZone said...

No. Pentium M is a P-III with some changes like a dedicated stack unit and micro-ops fusion. C2D is a step beyond Pentium M.

When dual core appeared, the prices did not double. Dual cores sell for about what single cores used to. Now quad core chips are again selling in this same price range. I believe by late 2007 AMD will be selling 2 quad core FX chips for the same price as 2 dual core FX chips today. I think Intel will have to follow suit.

Although doing 2 threads per core is possible I'm not seeing this in the short term. A dual socket quad core system allows 8 threads; it doesn't appear that 16 threads could be kept occupied.

Khalif, I could discuss decoding and threading but this would be long. This has to do with the balance among threads, decoders, and execution units. There are several approaches to this.

Ashenman, okay, if you want the real history then some of your perceptions are incorrect. K-6 was just as fast as P-II. What happened was that Intel used its monopoly power to derail 3DNow and replace it with SSE. K7 was beating P-III in 2000 because AMD got copper interconnects working a full year ahead of Intel. In 2001 P-III was falling behind and Williamette P4 was still shaky. If Intel had skipped Williamette they would have caught up in 2002 with a die shrink because they got the 130nm process under control before AMD did. They would not have had a clear lead like they did with Northwood which would probably mean that AMD would have reached 2003 with more market share in both desktop and servers (Athlon MP). However, this would mean that Intel would also have skipped Prescott and would have released an upgrade to P-III in late 2003 or early 2004. This would have skipped the bad associations with hot and power hungry Prescott cores. By itself I can't see that this would have changed the present. However, more money for AMD in 2002 would have meant that they could have started FAB 36 six months sooner so they would be further along now on 65nm. On the other hand, C2D would have been an easier step up from 2nd generation P-III so Intel might have released C2D a quarter sooner.

I think VR is wrong. I don't think Intel will release a new socket for X86 until 2009. I think in 2008 the new socket will only be for Itanium.

Red, as far as AMD's purchases go the history is like this: AMD disbanded its RISC team and used them to designed the RISC cored K5. AMD bought Nexgen and this became the K6. AMD bought technology from DEC and this became the K7. AMD tried to obtain process technology from Motorola and then UMC. They didn't get what they needed until the third try with IBM. AMD purchased Geode. And, now AMD has purchased ATI. The only big mistake they made was with Flash memory.

Streaming (GPU processing) is very similar to the classic Supercomputer vector architecture of which Cray and NEC are the only two survivors. I have no doubt that this will be the path beyond SSE.

Scientia from AMDZone said...

If THG truly cared about its reputation then they would use the PGI compiler rather than the Intel Compiler and they wouldn't compare overclocked Intel processors to stock speed AMD processors.

It is clear that THG does not care about its own reputation. Anandtech doesn't seem to care either. They proved this when they compared a dual socket Xeon system to a crippled Opteron system. I can't give them the benefit of the doubt when they make no effort to appear professional or objective.

ashenman said...

They may care about their reputation, but see no chance of it being tarnished by people who can't figure out they have no credibility.