another Dave Orton interview

[Bjorn]

Heh... How quickly we forget... nV30 had "more transistors" than R3x0, was manufactured on a smaller process, was clocked higher, and still ran much slower than R3x0, and suffered abysmal yields. Not FUD, fact.

I must say that i dont't really know what that has got to do with this. I thought that Dave O was talking about the NV4X.

Again, there was a big, big difference between what nVidia told us about the capabilities of nV30 and the actual chip itself, wasn't there? Not FUD, fact. What would make you assume before you know otherwise that nVidia was doing something different with nV40?

And again, we're not talking about the NV3X. And didn't i just say that i don't take Jen Hsuns words as gospel ? I'd rather wait until the cards are out on the market, have a couple of driver revisions and so forth..

He didn't say he was comparing clock speeds, he said he was surprised about the clock, seeing that both gpus are 16x1 and similar in other general ways. Again--I think he was talking obliquely about ATi's decision to go with low-k and how that affected ATi's clocks, contrasted with nVidia's decision not go with low-k, which definitely affects clocks and power consumption--else ATi wouldn't have used it. Not FUD, fact.

FUD = talking about Nvidias potential yield problems which afaik, he doesn't have any facts on. That he was surprised that a "60 million more transitors using a different manufacturing process" chip wasn't able to reach the same clockspeeds as the R420 was perhaps not FUD, just plain stupid imo. Stupid might be the wrong word, i should have said "pure marketing speech" cause it should be pretty obvious.

What Orton knows, however, is that the nV40 die size is bigger than R420 (fact), that it takes more power to operate reliably (fact), that it runs hotter at a lower MHz clock (fact.) These are all facts clearly in evidence at present. What would make you think otherwise?

I don't think otherwise.

OK, so what was it you think he doesn't know that's FUD? Didn't quite catch that... Additionally, it is a fact that R420-based 3d cards are shipping to retailers at the moment, but nV40-based products aren't (at least as far as any nV40 AIB OEMs or retail distributors have announced.) nV40 was announced first, right? Again, that's fact, not FUD.

That the NV40 was announced first hardly means that much since we don't know when the actual production started. And the battle over a new generation isn't fought in the first 3-4 weeks. It takes a bit longer then that.

Well, R360s in XTs shipped at 412MHz and were 8x1. R420PE will ship at 520MHz and is 16x1. That's not even counting all of the other >R360 capability, but just that by itself seems to me to be quite a fundamental, and substantial, "new feature" of R4x0 over R3x0. I think you are in such a rush to downplay R420 versus R360 that you literally can't see the forest for the tree...

Yes, the R420 has some new features over the R420. Yes, it's a good card. But it's basically a dual R360 with some minor tweaks and to me that's not "new" and i think it's a bit dissapointing.

 

[WaltC]

I must say that i dont't really know what that has got to do with this. I thought that Dave O was talking about the NV4X.

Heh... Try reading my response again in context with the statement of yours it was made in response to, in which you implied that since one chip had more transistors and was made on a different process than the other it was silly to compare clockspeeds. I was pointing out how nV30 had more transistors than R3x0, was built on a different process, was clocked higher than R3x0, but was slower than R3x0 and much inferior to R3x0, which seems to me to pretty much invalidate your objection to Orton's statement, doesn't it?

Like it or not, nV40 and R420 are competitors, and both are 16x1 gpus, and both employ hardware shaders dependent on MHz operation for certain aspects of their performance. I can't see how a clock-speed comment like Orton's is irrelevant under those circumstances.

And again, we're not talking about the NV3X. And didn't i just say that i don't take Jen Hsuns words as gospel ? I'd rather wait until the cards are out on the market, have a couple of driver revisions and so forth..

That's a convenient approach, but the fact is we are talking about statements made by nVidia--not the nV40 (which last time I checked has no voice)--and so previous statements nVidia's made about previous nVidia chips that turned out to be totally bogus certainly seem to me to be relevant to what nVidia's saying now about nV40. That's nVidia's fault, not mine, right? I've never once suggested nVidia should abandon credibility.

FUD = talking about Nvidias potential yield problems which afaik, he doesn't have any facts on. That he was surprised that a "60 million more transitors using a different manufacturing process" chip wasn't able to reach the same clockspeeds as the R420 was perhaps not FUD, just plain stupid imo. Stupid might be the wrong word, i should have said "pure marketing speech" cause it should be pretty obvious.

Again, I think it was far from stupid, and a pertinent remark on being surprised that nVidia voluntarily elected not to go low-k to obtain the benefits ATi has obtained from it in terms of clock and power consumption--which are two things crucial to 3d chips in terms of deployment and market penetration (other fundamental things like pixels per clock being equal, as they are in this case.)

I cannot imagine you calling JHH "stupid" if it was nV40 using low-k to go to higher clocks and he was contrasting ATi's deficiency in failing to do so. Were he able to do so, I believe he would do it with relish... (I mean, he's already talking about how dual, "optional" molex connectors make the 6800U an "advanced" product over primitive competitors who are crippled by offering only single-molex connector products...)

Again, it's not FUD that although nV40 was announced first, it will be second to market. If that tells *you* nothing about yields, then it's no wonder you don't understand the very simple thing Orton was saying there. In fact, I don't see how you could misunderstand it unless you simply wanted to.

Look, the biggest FUD of all is that fact that nobody knows when nV40-based products will ship, because retailers and nVidia's AIB OEMs aren't saying at the moment. I can't see how that's Orton's fault, or that ATi is responsible for that. R420 products are in the distribution pipeline right now--since nVidia announced ahead of ATi, then if yields were indeed great--OK, where are the products based on them? nVidia's self-inflicted FUD like that far exceeds anything Orton might ever say.

I don't think otherwise.

Well OK, then, much of what Orton said wasn't FUD at all, but rather fact, wasn't it? The die is bigger, it is clocked slower, and it does require more power and runs hotter.

That the NV40 was announced first hardly means that much since we don't know when the actual production started. And the battle over a new generation isn't fought in the first 3-4 weeks. It takes a bit longer then that.

Obviously, the fact that nV40 was announced first means nothing in terms of market availability... If it did, nV40 would already be shipping. If you want to say that nVidia is doing what they did with nV30 and announcing availability ahead of any firm knowledge as to nV40 yields--then I won't argue with you, because I don't know. I'll just remind you that in your previous post you quoted nVidia talking about how much better the yields are--so how can they *know* that without already being deep into the production process? And if they are already producing, where are the nV40 products?

I mean, it's a bit strange, isn't it, when nVidia says, "yields are great" but the products based on those great-yielding chips are nowhere to be found? Sure seems contradictory to me. Bottom line is: shipping the product is everything.

Yes, the R420 has some new features over the R420. Yes, it's a good card. But it's basically a dual R360 with some minor tweaks and to me that's not "new" and i think it's a bit dissapointing.

You are centainly entitled to your opinion, of course...

 

[Bjorn]

Like it or not, nV40 and R420 are competitors, and both are 16x1 gpus, and both employ hardware shaders dependent on MHz operation for certain aspects of their performance. I can't see how a clock-speed comment like Orton's is irrelevant under those circumstances.

Orton's comments are only valid if both cards have the same featureset. You might be of the opinion that they do, i don't.

I cannot imagine you calling JHH "stupid" if it was nV40 using low-k to go to higher clocks and he was contrasting ATi's deficiency in failing to do so. Were he able to do so, I believe he would do it with relish... (I mean, he's already talking about how dual, "optional" molex connectors make the 6800U an "advanced" product over primitive competitors who are crippled by offering only single-molex connector products...)

The R9700 had lower clockspeeds then the NV3X and was still a much better card (not connected to the NV4X vs R420, just an example ). If Jen Hsun hade made comments (perhaps he did, i don't know) about how he couldn't understand why they couldn't achieve higher clockrates back then, then yes, that would have been pretty stupid.

 

[Dave Baumann]

Bjorn, If you want to take those comments from the CC and disect them, here is a transcript of what was actually stated - this are the two Q/A's that are pertinent to the PSU issue

<Q – David Wu>: Can you elaborate one more point? Apparently to get to the Ultra – 6800 Ultra, you need an extra power plug in there. You told me essentially Shader 3.0 feature proves your customers' purchasing decision, which is a good one. But if they’re to buy an extra power plug, isn’t that a negative?

< Jen-Hsun Huang>: It is but I think everybody just got it wrong and there is, you know, the – some of the plug that’s going around, it's just a bit of nonsense. But basically the way that it works is this. We designed GeForce 6800 Ultra to have an optional, and the operating word in optional, really important word – optional, overclocking capability. When you overclock these devices, you are going to run out of power in your system. it kind of makes sense, right?

<Q – Andrew Root>: Hi. Thanks very much. Actually, Jen-Hsun, I just wanted a follow-up on the power issue that you were going though which is very interesting actually; kind of controversial as you know. If you were to look at the power supply that the system needs whether or not they plug-in the additional connector, do you still need a more substantial power supply in the board or that's also a decision the board maker can make? And then just secondarily, the benchmarking studies that we've been reading with very nice reviews, the NV40 product. Is that including the second power supply plugged in or is that using the basic clocking?

< Jen-Hsun Huang>: So, first of all, whether you plug in one power supply or two, we ship all of our GPUs at all exactly the same speed. You know, I think the benchmarking recently web-reviews, I understand our competitors can pick boards of all different speeds and every single review had a different speed clock. We don’t do that. Every single one of our GPUs are shipped with the same speed. And the reason for that is because consumers need to know what they are buying and we believe that if we position the GeForce 6800 Ultra at a particular speed, that ought to be the speed that the end consumer ought to know that, with a great deal of confidence, they’re going get, if not more. So that’s number one. Number two, in terms of the reviewers, some of them had both power supplies plugged in and some of them only had one power cord plugged in. One power cord gives you about 350 watts and if you would like to have an optional second power plug with a 480 watt power supply, that’s terrific. What all of the reviewers with one power connector was able to reach the standard frequency we ship at for GeForce 6800 Ultra. Some of them overclocked our GPU and some of them were able to achieve, you know, quite good headroom. And the reason for that is because we are shipping at a pretty comfortable frequency and some people have achieved as much as 50 megahertz in just being able to overclock their GPU. Whereas, you know, I think some comments were made about not being able to get another 1 or 2 megahertz out of the competition's board which kind of makes sense because they were hand-picked. And so, the bottom line for you is that if you ship with one power – if you only connect up one power supply, you ought to be able to achieve our GeForce 6800 Ultra specs. If you would like to overclock it, we’d recommend that you put in another power supply.

So, first he calls the issue a bit of nonsense thats going around - this is nonsense that NVIDIA's reviewer guides tell us. He then tells us that the 6800 Ultra was designed with an optional Power socket (note the past tense - if it was designed in such a fashion why haven't we seen it yet; he didn't state that it will be designed in that fashion (seeing as the boards that have already gone out clearly weren't designed in such a fashion, suggesting that they are still to design them in such a fashion yet).

He further states that some reviewers had one Power socket plugged in and other didn't - although I've not followed up every review, I haven't seen a case where a reviewer tried it with successful operation; indeed it would go against their own reviewers guide and further testing on the "reviewers boards" we've seen suggests that normal operation can't be achieved (again, contradicting JHH's words as he states "What all of the reviewers with one power connector was able to reach the standard frequency we ship at for GeForce 6800 Ultra"). He also states that one power cord can supply about 350W - I think this is a little much for one cord.

[Interestingly he also states that everyone of their boards will ship at the same speed, despite the 450MHz Ultras - not Ultra Extreme's according to UK PR - going to certain select vendors]

So, IMO, I would look at those words and try and reconsile them with the evidence of what is out there at the moment and ask if there isn't just some analyst bluffing going on there.

 

[Sabastian]

Regarding shipping. It was ATi management (KY Ho) that announced a couple of years ago that ATi would not announce a product without being able to ship it. nVidia never made such a decree. Anyhow I think that the shipping is simply a matter of ATi being a little more prepared for the launch it is their modus operandi if you like and X800 vpu's are shipping as we speak. I don't know about the rumors concerning the 6800 U. There were rumors that NV dropped IBM all together for fab and moved to TSMC and that TSMC had not produced a single 6800 U as of yet. But we hear NVs CEO make claims that they will ship the 6800 U on first silicon. Rumors are rumors and CEOs are CEOs. Until we actually see the 6800 U available on mass (not in limited quantities) then we don't know what the matter is regarding it. Certainly NVs CEO dodging the question of the products availability over the next couple of months was telling though. Having silicon in mass avialability first is an advantage. It allows for thick margins without worries of competitive pressures. It also helps to garner market mindshare. I really think NV has an uphill battle here.

 

[WaltC]

Orton's comments are only valid if both cards have the same featureset. You might be of the opinion that they do, i don't.

No, I'm of the opinion that Orton's comments are entirely valid since nVidia doesn't *have* a gpu with the "same feature set" to field against R420. What they have is all they have to compare, and so any and all comparisons are relevant since both gpus are in the same general class of product and the same price range. Hence, direct comparisons between them on many levels are inevtiable, and Orton will not be the only one making them. (Everyone else notices that while both nV40U and R420PE are made on the same .13 micron manufacturing process there's a significant clock-rate disparity between them, but you seem to think only Orton is "stupid" for saying so...)

What you seem to be saying is that in your opinion nV40 and R420 should *not be compared* since nVidia *says* nV40 includes effective support for ps3.0 and things like fp frame-buffer blending. I think you'll find you are in the distinct minority of opinion on that point, and that most people will think the idea of not comparing the two directly is an absurdity...

 

[Bjorn]

So, IMO, I would look at those words and try and reconsile them with the evidence of what is out there at the moment and ask if there isn't just some analyst bluffing going on there.

You never know. I sure as hell don't believe it until i see reviews of actual cards but i like to keep an open mind until then.

 

[Bjorn]

What you seem to be saying is that in your opinion nV40 and R420 should *not be compared* since nVidia *says* nV40 includes effective support for ps3.0 and things like fp frame-buffer blending. I think you'll find you are in the distinct minority of opinion on that point, and that most people will think the idea of not comparing the two directly is an absurdity...

No, that's very far from what i'm saying. There is a difference between comparing them afa benchmarks and general usage goes and "transistor count/process vs clock frequency". But i'm repeating myself here so i think we just have to disagree.

 

[Dave Baumann]

Yes but before they were done in 1 site, now and till R800 it will be more or less 2 sites, and beginning with R800 all the 3 sites will be completly involved.

I would partially go along with Sabastians lin of thinking here as he states the engineering is already unified and the R600 will already have been worked on by all 3, its just that R800 will be more unified.

There are also "bad rumors" about ATI as well. That the XT PE is a "cherry pick" and they're having problems with XT yields. I don't believe either way, but these things flow into my mailbox.

Availability of both the Ultra and XT are most likely being dictated by the availability of RAM at the moment more than anything else. Avialability of the PRO doesn't really suggest much in the way of yield issues, although thats not at the full spec - I've heard some things about yields of R420 that will negatively impact some of the X800 line, but thats likely to hit the SE edition most...

FUD = talking about Nvidias potential yield problems which afaik, he doesn't have any facts on. That he was surprised that a "60 million more transitors using a different manufacturing process" chip wasn't able to reach the same clockspeeds as the R420 was perhaps not FUD, just plain stupid imo. Stupid might be the wrong word, i should have said "pure marketing speech" cause it should be pretty obvious.

I took that as being more tongue in cheek commentry myself, just to rib the guys at NVIDIA that he knows and plays golf with every Wednesday... It would be interesting to know if the actual content of that interview was taken before or after NVIDIA's CC.

 

[WaltC]

The R9700 had lower clockspeeds then the NV3X and was still a much better card (not connected to the NV4X vs R420, just an example ). If Jen Hsun hade made comments (perhaps he did, i don't know) about how he couldn't understand why they couldn't achieve higher clockrates back then, then yes, that would have been pretty stupid.

Sigh...again, unlike R3x0/nV3x, nV40/R420 are both manufactured at .13 microns, and unlike R3x0/nV3x, both nV40 and R420 have 16x1 pixel pipe organizations. So, there is a much greater probability that--other things being equal--the clocks for both should have been closer than they were. I think that's all Orton was saying there, which seems entirely logical to me. What's different is not only the transistor count (how different that actually is we don't really know since ATi & nVidia count them differently), and the fact that R420 is made under low-k and nV40 is not seems to be of greatest import here.

I remember that during the time nVidia was stuggling to get nV30 out of the door there were many rumors going around about the nV30 delays being caused by a failure to implement low-k into nV30 production. Now here we are with a low-k-less nV40 as well. I'd really like to know what there is about the nV3x/nV4x layout that seems to preclude nVidia utilizing low-k as ATi has been able to do. nVidia has access to the same FABs that ATi uses, so that cannot be it. I wonder what else it might be?

 

[WaltC]

I took that as being more tongue in cheek commentry myself, just to rib the guys at NVIDIA that he knows and plays golf with every Wednesday... It would be interesting to know if the actual content of that interview was taken before or after NVIDIA's CC.

Yes, I mean in your interview Orton directly stated it was "tounge 'n cheek," didn't he?... I took it to mean he didn't know for certain about nV40 yields, but that he had his suspicions.

 

[Evildeus]

Yes but before they were done in 1 site, now and till R800 it will be more or less 2 sites, and beginning with R800 all the 3 sites will be completly involved.

I would partially go along with Sabastians lin of thinking here as he states the engineering is already unified and the R600 will already have been worked on by all 3, its just that R800 will be more unified.

Well, it all depends on what you mean by unified. A chip being made in 2 sites, is more than 1 site, a chip in 2 site plus a bit in another site is more than just 2 sites. But from what he said, it seems that the R800 will be the first one to be done in the 3 sites all working together.

 

[Bjorn]

Sigh...again, unlike R3x0/nV3x, nV40/R420 are both manufactured at .13 microns, and unlike R3x0/nV3x, both nV40 and R420 have 16x1 pixel pipe organizations. So, there is a much greater probability that--other things being equal--the clocks for both should have been closer than they were.

You're not the only one who's sighing. The thing is, i think the "Other things being equal" is an invalid statement. You must disagree with that and that's fine by me. We disagree.

 

[Bjorn]

Yes, I mean in your interview Orton directly stated it was "tounge 'n cheek," didn't he?... I took it to mean he didn't know for certain about nV40 yields, but that he had his suspicions.

So Jen Hsun's talk about Ati handpicking boards for reviewers and such crap wasn't FUD either then, or ?

I mean, he don't know for certain but he had his suspicions.

 

[Dave Baumann]

If you dissect in, for example, to the R600 product, with is our next, next generation, that development team is all three sites - Orlando, Silicon Valley, Marlborough – but the architectural centre team is in the Valley, as you point out, but all three are part of that organisation.

 

[Evildeus]

May i re-quote myself?

Sabastian,
Here's the part that to me seems to say that R800 is the really beginning of 3 unified teams:

So the R600 family will mainly be centred primarily in the Valley and Orlando with a little bit from Marlborough, and then the R800 would be more unified.

http://www.beyond3d.com/interviews/daveorton/index.php?p=3

Sure, R600 will be done on 3 sites, but the 3 sites will not be involved at the same degree before R800.

 

[Dave Baumann]

Ergo, they are already working on a single, unified design.

 

[Sabastian]

May i re-quote myself?

Sabastian,
Here's the part that to me seems to say that R800 is the really beginning of 3 unified teams:

So the R600 family will mainly be centred primarily in the Valley and Orlando with a little bit from Marlborough, and then the R800 would be more unified.

http://www.beyond3d.com/interviews/daveorton/index.php?p=3

Sure, R600 will be done on 3 sites, but the 3 sites will not be involved at the same degree before R800.

Dear god man.. Ok fine. But the situation is not the way it was like for example when the Marlborough team designed the R300... agreed?

 

[Evildeus]

I don't really understand your point, your are saying that it's already done on 3 sites for R600, i'm saying that R800 will be the first true unified chip with te 3 sites working on hand in hand together with the same workload dispatched.

From my little experience in organization consulting, only when the 3 sites are totally operational, we can talk of unified, before, it's just adjustment on the path to achieve the goal.

 

[Evildeus]

Dear god man.. Ok fine. But the situation is not the way it was like for example when the Marlborough team designed the R300... agreed?

Did i say otherwise? Sorry for being precise