another Dave Orton interview

[Joe DeFuria]

Yes, i don't take DO's answers as gospel because he'll say whatever is in Ati's best interest which is not necessarily the truth (or at least not the whole truth).

He admitted the restructuting of their design teams because of execution issues. In other words, when mistakes in approach were made, he seems to have no qualms of fessing up to them. I see no reason to not believe him.

Now, just because he says that he believes he COULD have done a PS 3.0 part for R420, doesn't mean they would have been successful...but I see no reason to doubt that they made their decision based on bean counting, not on what they thought their teams were technically capable of.

 

[Bjorn]

Now, just because he says that he believes he COULD have done a PS 3.0 part for R420, doesn't mean they would have been successful...but I see no reason to doubt that they made their decision based on bean counting, not on what they thought their teams were technically capable of.

I have absolutely no doubt that they are capable of doing a SM3.0 part. None what so ever. I'm only guessing that since the R420 has no new (close to ) features, Ati decided to put their resources on upcoming projects instead. And this goes hand in hand with the bean counting because of the reasons i mentioned before (less driver work, less transistors/performance because of less features and FP24 -> cheaper to produce, although that's pretty obvious i guess).

 

[Bjorn]

He admitted the restructuting of their design teams because of execution issues. In other words, when mistakes in approach were made, he seems to have no qualms of fessing up to them.

That's true. But what was the cause of the restructuring and execution issues ?

As you can see in this thread, i'm of the opinion that the the cause is the X-Box 2 and upcoming Nintendo chips. And possibly the R500. And that Ati didn't simply decide that they didn't want to do a SM3.0 part. More like they had to go for a R300 on steroids. The good thing is that Ati is in the position that they can get away with this, Nvidia wasn't with the GF4 -> NV3X.

 

[Joe DeFuria]

He admitted the restructuting of their design teams because of execution issues. In other words, when mistakes in approach were made, he seems to have no qualms of fessing up to them.

That's true. But what was the cause of the restructuring and execution issues ?

Didn't he answer that question as well? Primarily, that it's tough to "re-use" technology (information sharing) when your teams are isolated.

As you can see in this thread, i'm of the opinion that the the cause is the X-Box 2 and upcoming Nintendo chips. And possibly the R500. And that Ati didn't simply decide that they didn't want to do a SM3.0 part.

You seem to be taking all of these things in isolatation. Surely...they all play a roll in the decision.

However, you are saying that if ATI didn't have x-box / Nintendo, then they would have SM 3.0. I don't see why they wouldn't have made the same decision.

More like they had to go for a R300 on steroids. And Ati is in the position that they can get away with this, Nvidia wasn't with the GF4 -> NV3X.

Well, yes. If the R300 core sucked, they couldn't get away with it. This has nothing to do with XBox / Nintendo. It has to do with having a solid core in the R3xx, making it feasible to "get away with it". (This is why nVidia went with NV40...because they had to.) The R420 as we know it is the cheapest and least risky way for ATI to compete effectively this round from a development and productization standpoint, and having x-box or not would not change that fact.

 

[Bjorn]

You seem to be taking all of these things in isolatation. Surely...they all play a roll in the decision.

Of course. All i'm saying is that it's not as simple as "we didn't want to". And then we had some (imo) bad excuses as for why they didn't add any new features:

What we felt was that in order to really appeal to the developers who are shipping volume games in ’04 Shader 2.0 would be the volume shader model of use.

How on earth would that appeal to the developers ? It's pretty obvious that the 04 shader model will be 2.0 but the developers already have 9800 XT's for that. And i'm guessing that most developers will switch rather quickly to the NV4X to be ready for the upcoming next gen consoles (take the Unreal Engine 3 f.e). (But as always developers are one thing, consumers another).

However, you are saying that if ATI didn't have x-box / Nintendo, then they would have SM 3.0. I don't see why they wouldn't have made the same decision.

Not necessarily no, but i happen to think that we would have seen a different chip then. Possibly not a SM3.0 part but a lot more new features for sure.

 

[Joe DeFuria]

What we felt was that in order to really appeal to the developers who are shipping volume games in ’04 Shader 2.0 would be the volume shader model of use.

How on earth would that appeal to the developers ?

Because you're concentrating on getting them the performance for features they're developing for. Seriously, isn't it obvious? ATI is saying the decided to stick with a certain die size. If they do that, and also add in 3.0...something else is going to give. Namely: performance. ATI is saying that given a choice, devs at this time would rather see their 2004 titles running well, than have more features to play with and not so large a performance increase for shipping and near shipping titles (forcing a toning down of shader 2.0 effects to get performance.)

Of course, you can try to do BOTH (and NV40 is an example of this), but then, you have to go to a larger die, risk yields, profit margins, and actually having a producible part.

 

[Bjorn]

Because you're concentrating on getting them the performance for features they're developing for. Seriously, isn't it obvious?

What i meant is that developers with games coming out this year will obviously be more worried about the 9500+ series then the by comparision minor number of X800's that's going to be released this year.
Especially since games coming out this year will probably have their feature set pretty much set in stone. I doubt that they'll add any new features just because of the X800, other then 3Dc support that is but that's doable on the R300 series also (through DXT5).

Their games will run faster with the X800 though and the people that buys the X800 will of course be happy about that. Which you can say is good for the developers but that's another thing imo. And the NV4X is obviously helping the developers in just the same way, but is also offering SM3.0, FP blending and other additional features which is going to be good to have when adapting your engine for the X-Box 2.

ATI is saying the decided to stick with a certain die size. If they do that, and also add in 3.0...something else is going to give. Namely: performance.

A native 12 pipe X800 with SM3.0 (or at least VS3.0, FP blending which seems to be the things that develoers like the most) running at 500+ MHz would have been a very good card compared to the 9800 and i doubt that it would have needed that many more transistors then the current X800.

Of course, you can try to do BOTH (and NV40 is an example of this), but then, you have to go to a larger die, risk yields, profit margins, and actually having a producible part.

That's true. And it remains to be seen how all this pans out for the 6800.

 

[Heathen]

i doubt that it would have needed that many more transistors then the current X800.

Which X800?

 

[Joe DeFuria]

What i meant is that developers with games coming out this year will obviously be more worried about the 9500+ series then the by comparision minor number of X800's that's going to be released this year.

Actually, they're probably really worried about the 5200+ series...but that's really another debate.

Especially since games coming out this year will probably have their feature set pretty much set in stone. I doubt that they'll add any new features just because of the X800, other then 3Dc support that is but that's doable on the R300 series also (through DXT5).

That's jsut it...most feature sets will be pretty much set in stone for titles this year, the better the game performs, the better the dev "looks". The dev can more easily exploit (make heavier use of) existing features in their engines for higher performing cards. Instead of a few normal maps on a couple characters....do it everywhere for example.

A native 12 pipe X800 with SM3.0 (or at least VS3.0, FP blending which seems to be the things that develoers like the most) running at 500+ MHz would have been a very good card compared to the 9800 and i doubt that it would have needed that many more transistors then the current X800.

That's certainly not the card that I want. Again, you seem to be overlooking the aspect of higher performing parts, allows a higher exploitation (better quality effects) of them.

It's not just new "features" that allows for more quality effects...its better performance of existing features.

Doom3 is the classic example..using tech available since DX7 era, but will probably only really get exploited with DX9 level cards....most because of performance.

And it remains to be seen how all this pans out for the 6800.

Well already know that:
1) Its larger
2) It's more power consumption.
3) It's overall slower. (I believe that's the concensus, correct?)

What we don't know is how "producible" it is in comparison, but we have one big clue: The X800 Pro is shipping, and the "lower end" version of the NV40 isn't. The fact that ATI has the more voluminous part shipping before their high end part is pretty telling IMO. One of the reviews mentioned that the only "reason" why the XT is later, is because they're waiting on the 600 Mhz Ram shipments. That could also explain the lack of the presence of the 6800 Ultra (assuming they aren't actually going to ship with overclocked Ram), but does not explain the absense of the 6800 non ultra or GT.

 

[Bouncing Zabaglione Bros.]

You could of course turn that around and say that most people were surprised that the R420 has close to zero new features. It was clear pretty early that it wasn't going to be a SM3.0 part but that it didn't have VS3.0 or any tweaks to FSAA/AF or things like that was not what i expected.

Well it was a completely loaded question, so an answer like this is hardly suprising.

 

[Bouncing Zabaglione Bros.]

The final cards aren't out on the market though and the 6800 GT will have only one molex connector if i'm not mistaken. Shouldn't be a inelegant solution for the consumer in any way.

But that's not the same product. You get a much lower level of performance.

 

[Sabastian]

The final cards aren't out on the market though and the 6800 GT will have only one molex connector if i'm not mistaken. Shouldn't be a inelegant solution for the consumer in any way.

But that's not the same product. You get a much lower level of performance.

While it may not be the same product it is the same core only down clocked but still retaining the 16 pipeline design. I think with the right pricing the GT may end up a great seller for NV but with weaker margins.

Bjorn might have a point regarding the change in the roadmap as a result of the xbox2. But there is no doubt in my mind that ATi has the technical skills needed to implement the PS3.0 speck. Sure ATi might have changed the road map so that they could make additional efforts towards the xbox2 architecture because they wanted too not because they had too. Further ATi may have stolen a page from the NV2x roadmap. The GF4 series of cards were an unmitigated success features took a back seat to speed. But this situation is not entirely analogous with the GF3-4 series at all. First off the speed disparities between the Radeon 8500 and the GF4Ti 4600 seemed to be more pronounced at the time then the disparities between the X800XT vs 6800 U. Secondly the X800XT is more then simply a speed bump, its a massive doubling of the Radeon 9800 in terms of performance.

I don't think PS3.0 will do much for NV in the way of sales to end users. Sure developers will want it but it isn't the end all beat all of features where the end user is concerned. 3Dc may be more of a going concern in the immediate future for gamers/enthusiast. Finally normal maps are becoming a going concern for developers. NV does have a bright spot concerning their new core though... mostly that is performance. While we don't know what or how PS3.0 does for performance we do know that on lower resolutions with 4xAA teh 6800 U performs at least as fast as the X800XT. (don't turn up the res or AA beyond 4XAA though) But until NV can make it available they are really at a disadvantage. No one is going to upgrade (if you could call it that.) from an R420 based core to a NV4x based one. NV has really put them selves at a disadvantage regarding availability particularly if you consider how finite the enthusiast market is. NV better get them cards on the market asap IMO else ATi will gobble the lions share of the enthusiast market but that isn't a guarantee for success ether, ATi is very popular these days.

Regarding "elegance".. I think NVs core design with PS3.0/222 million transistors is quite an achievement actually.(elegant core design? can't say really not a chip engineer.) The R420 is more a kin to a small but mighty pit bull on steroids ..what is elegant about that? Beyond that though looking at teh cards the 6800U is analogous to a Cadillac (size power consumption ect) this is not an elegant looking/acting card but its got more stuff. While the R420 in terms of it size is more like a Porsche and this is where the label of "elegant" might be appropriate.

 

[Dave Baumann]

As Joe points out, answers in that interview weren't ones that appear to be digding the issues - he's admitted that they have restructured their engineering because what they initially planned wasn't going to work and he's admitted that they are not afrid to change the roadmap when they make mistakes. When I asked about SM3.0, to which his reply was "we don't think developers are really focusing on this" my followup question was tantamount to asking "Are you bullshitting me - isn't more a case you just don't have something on the table at the moment?" to which his answer was (paraphrasing) "We could have done it, but we really wouldn't mind making some money off these chips, thanks. We'll wait for 90nm so we can still with SM3.0". Given the frankness of his other replies I don't see that he was dodging this particular one.

Note, I think some of the things that went on in this interview are subject to change, and I've even heard that thing might have changed already so it wouldn't surprise me if ATI do some things that will go against the grain of some of these responses in the future.

(BTW - what I took away from this interview the most is that, as yet, we've yet to see the full potential of ATI's engineering working on a single project)

 

[Joe DeFuria]

(BTW - what I took away from this interview the most is that, as yet, we've yet to see the full potential of ATI's engineering working on a single project)

Yup, though unfortiunately it looks like we'll have to wait until R800 before we really find out!

 

[Sabastian]

(BTW - what I took away from this interview the most is that, as yet, we've yet to see the full potential of ATI's engineering working on a single project)

That is interesting. ATI has a massive and talented engineering core. I think now that ATi has managed some real inroads in the PC market over the last couple of years and their bottom line is showing as much they are going to redouble their efforts on their core market and have less focus on other markets.(HDTV for one.) Which project will the entire team be working on though is the question. I am of the mind it will be a multi pronged effort all centered around the xbox2 vpu... and its PC derivatives somehow. IMO the xbox 2 project and its PC derivatives is ATIs #1 priority at this point, this is likely what they are focused on as one unit. It is primary that they don't miss a beat with that project diverting so much attention away from their core business so they have to make it pay off in the PC market of which I am sure they will.

EDIT: The R420 most certianly enabled that too happen.

 

[Sabastian]

Yup, though unfortiunately it looks like we'll have to wait until R800 before we really find out!

I disagree of course.

 

[Evildeus]

Yup, though unfortiunately it looks like we'll have to wait until R800 before we really find out!

I disagree of course.

But that's what he said in DB interview

 

[Sabastian]

Yup, though unfortiunately it looks like we'll have to wait until R800 before we really find out!

I disagree of course.

But that's what he said in DB interview

I didn't get that sense. I must be reading this wrong then?

What we are now moving towards is actually a unified design team of both east and west coast, that will develop our next generations of platforms, from R300 to R400 to R500 to R600 to R700, instead of a ping-pong ball between them both. Within that one organisation we need to think about where do we architecturally innovate and where do we not in order to hit the right development cycles to keep the leadership, but it will be one organisation.

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.

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

 

[McDusty]

(BTW - what I took away from this interview the most is that, as yet, we've yet to see the full potential of ATI's engineering working on a single project)

True, but to many cooks can spoil the broth!

 

[Sabastian]

(BTW - what I took away from this interview the most is that, as yet, we've yet to see the full potential of ATI's engineering working on a single project)

True, but to many cooks can spoil the broth!

heh, speaking from experience... that is only true if they don't operate in a structured approach. A lot of cooks can make for one fast/good stew.