AMD: R8xx Speculation

[Sound_Card]

Not sure what metric you can use to come to the conclusion that their execution has been solid. They don't have anything to show for it.

Not really sure what G92/G94 have to do with anything. Are you implying that future Nvidia products will inevitably be less competitive than G92/G94?

Yes, Nvidia lost money but that's not a reflection on anything so simple as manufacturing costs of large die GPUs. There are huge number of other significant factors including the amount of R&D money being pumped into high-margin businesses. Compare the Q1 R&D numbers - Nvidia 211m, AMD 305m. What portion of that 305m would you guess is attributable to GPUs?

What is wrong with their execution? It's been pretty good as of late. I'm confused more so then you.

I bring up G92 because they brought that down to combat RV770pro and it worked/s fairly well. Are they going to move a cut down GT200 down as well when RV870 hits? As far as I can tell, they have no such GPU (like G92) for GT200 series that can scale around this problem. Major reason being less relevant then before is looking at the technology feature sets these current GPU have and looking forward with an assumption. RV870 is going to be DX11, and I can't possibly fathom NV using a rebaged (more likely) DX10 GPU to compete and get away with it as easy this time. Again, I assuming GT300 is another mammoth.

 

[trinibwoy]

What is wrong with their execution? It's been pretty good as of late. I'm confused more so then you.

Just wondering what tangible results you're looking at to draw that conclusion.

I bring up G92 because they brought that down to combat RV770pro and it worked/s fairly well. Are they going to move a cut down GT200 down as well when RV870 hits?

Who knows. I wouldn't put it past them but there are too many unknowns at this point on both sides. It could be another RV670 vs G94 or RV770 vs GT200 (more likely).

 

[Silent_Buddha]

Not really sure what G92/G94 have to do with anything. Are you implying that future Nvidia products will inevitably be less competitive than G92/G94?

G92/G94 has everything to do with it. If Nvidia had to rely only on their new cards they would be losing quite badly. G92(b) has been able to keep Nvidia marketshare where it is and even regained some lost ground from Q3 08. Otherwise, ATI would have not only maintained the ground they gained in Q3 08, but most likely expanded upon it.

Even with price adjustments Rv770 is still outselling GT200.

The fact that G9x continues to be Nvidia's best selling chip is either a testament to how good that was or how bad GT200 is.

As well, after writing off inventories in Q4 08, Nvidia is free to price that existing inventory however they wish as it is almost pure profit from that point forward until they run out of that inventory.

That allows them to continue to keep that chip relevant.

Coupled with actively lobbying against Dx10.1 features (oddly enough they'll have to starty lobbying/helping devs with those features in Dx11) helped prevent too many Dx10.1 games from reaching the market and showing their current chips in an even worse light.

Which speaks volumes for the benefits of having had market dominance for so long.

Considering the environment Rv770 had to work in, I think ATI executed quite well. And certainly executed far better than Nvidia did with GT200. Of course, the argument with GT200 is that the focus was mainly on CUDA and thus there may eventually be dividends to reap from that with OpenCL, but we'll have to wait a bit more before being to determine the impact of those design decisions.

Regards,
SB

 

[LunchBox]

IMHO... don't kill me

I think what ATi/AMD really needs to do is offer game developers the same support Nvidia is giving so that their technology will shine in games. I will probably get chased with pitchforks for saying this but, IMHO, one of the reasons Nvidia has over ATI/AMD is games like Crysis runs better on Nvidia compared to ATI/AMD because it was coded optimally to run on Nvidia, not because Nvidia has the "better" hardware. Driver "hotfixes" can only do so much...

As far as their Latest chip is concerned, it seems that the upgrade is a little modest compared to their last chip. maybe because I was so surprised to see their ALU count jump so high in their last one that I am expecting them to do the same if not more for their new chip.

 

[trinibwoy]

G92/G94 has everything to do with it. If Nvidia had to rely only on their new cards they would be losing quite badly.

Lol what? In what way is G92b "old" compared to GT200? Are you suggesting GT200 should be sold for $99?

Even with price adjustments Rv770 is still outselling GT200.

Really - based on what? JHH claimed all their market share gains were in the high-end.

The fact that G9x continues to be Nvidia's best selling chip is either a testament to how good that was or how bad GT200 is.

Why the artificial separation between G9x and GT200? They're both Nvidia products aimed at different price levels. Or is selling G9x "unfair" in some way now? In any case, all this doesn't answer the question of why G92 is relevant for the upcoming battle.

Considering the environment Rv770 had to work in, I think ATI executed quite well.

Why is today's environment especially hard for ATI compared to previous periods? The question still stands - what've they got to show for their efforts so far?

 

[Lukfi]

=>LunchBox: I regard that as a very legitimate opinion. The interesting thing is that ATI does actually work with developers like nVidia does, although at a much smaller scale. Mainly because it costs lots of money.

 

[Jawed]

Really - based on what? JHH claimed all their market share gains were in the high-end.

Doesn't say anything about quantity of sales though. In the simplest interpretation that merely means that GTX295 + GTX280/285 outsold HD4870X2.

I don't know how we could compare all sales of RV770 and GT200. Steam data seems to be in contradiction of the Mercury data, so there's no useful breakdown out there.

The ultimate measure of a GPU's effectiveness, in isolation, is return on investment. AMD's clearly way behind NVidia on the services side of GPU (developer relations, professional services, CUDA assistance) so there's a big chunk of value-add investment that AMD's leaving on the table/can't afford, which affects medium-/long-term market share.

The old ATI philosophy of what seemed to be "we just have to build a better product and it'll sell itself" seems dominant (even if there was some devrel). Intel and NVidia both do a lot of services as far as I can tell, while AMD seems to have long had a "minimal services and no marketing" approach. I think AMD's smelt the coffee as services are starting to have a noticeable effect and there is actual AMD marketing out there, but it's a long road.

Jawed

 

[Rangers]

Trinibwoy, what data do you have that AMD lost share from the HD2900 days?

It may not be scientific, but if you look at message board posts (various boards of course, such as [H]), newegg reviews (number of reviews for a card seems to correlate with purchases, there were almost no reviews for AMd cards in the 2900 days, now they are up there with the # of reviews for Nvidia cards), Steam survey, etc, all seem to show major gains for AMD since the 4800 series debuts. This makes a lot of sense as well since they are solid performers for a good price, where HD2900 was not.

Are you sure the market data you're looking at isn't including IGP? I really have a hard time believing AMD hasnt gained major share in actual purchasers of mid range and up discrete cards. I really refuse to believe that. Now, maybe Nvidia has all the OEM contract for discrete cards locked down ( or whatever you would call cards contained in prebuilt systems), as far as that side goes. But there's no way I believe end user purchases of discrete cards haven't made major gains toward AMD this time around. No way. Now, I fully believe Nvidia has at least 50% of that market, but I'd wager in the R600 era they had 80%+ of it. I mean NOBODY was buying ATI cards back then.

Beyond that I dont know what to say, if people refuse to buy AMD cards simply because they are AMD, as you seem to suggest, well then it doesnt matter what AMD does so theres no point in them trying at all anyway, thus your strategy debate is irrelevant.

It is obvious Nvidia has put out a competitive product to AMD at every price point. Case in point, the GTX275 they introduced to compete with HD4890 is almost the exact same performance for the exact same price. Therefore if you like Nvidia cards, you almost always have an equal Nvidia option at any price point. Of course we all know the huge die size differences there which favor AMD.

Personally I figure the reason AMD doesnt cut prices more and really hurt Nvidia, as it seems they could, is simply that the CPU side is struggling so much they need every ounce of profit they can get get from the GPU side within reason. Realistically AMD CPU's dragging down the GPU's, it's kind of a nightmare for ATI.

Now, I guess if you're saying AMD needs to go for THE clear fastest card, period, then I'd have to agree with you on that. However, they are already pretty close with 4890. Theyre maybe 10% from GTX 285 now, once you OC a 4890 it's typically faster than a 285. And the dual GPU picture gets even murkier who is fastest. And supposedly theyre not even trying to be fastest, yet they compete with Nvidias top GPU as is. So again, I dont know what AMD has to do, but theyve clearly out-engineered Nvidia this go round (just as Nvidia had out engineered them in the last go round). I mean, it seems unlikely to me that AMD is ever going to me more than 10-20% faster than Nvidia, in general, as these companies are typically so competitive. And theyre within 10% now so..so basically if you're saying AMD needs to be 50% faster than Nvidia to start gaining major share, is that realistic?

 

[Ailuros]

I admit having a split personality; meaning I'm answering my own questions:

Do I believe RV870 will be an outstanding D3D11 performance GPU?

Yes.

Is AMD going to retain for the next generation to the same degree their advantages when comparing RV7x0 to GT200?

No.

Is G92 outdated compared to GT200?

Both are D3D10 GPUs.

Personally I would have preferred AMD not to sell their handheld division and to concentrate even more aggressively on professional markets. IMHLO ATI has been limited way too much to something I'd call the "middle playground". I hope things will change in the less foreseeable future when this darned financial crysis is over and AMD could be able to divert consistently more and more resources to its graphics division in order for it to grow over the years.

 

[Sound_Card]

Trinibwoy, what data do you have that AMD lost share from the HD2900 days?

It may not be scientific, but if you look at message board posts (various boards of course, such as [H]), newegg reviews (number of reviews for a card seems to correlate with purchases, there were almost no reviews for AMd cards in the 2900 days, now they are up there with the # of reviews for Nvidia cards), Steam survey, etc, all seem to show major gains for AMD since the 4800 series debuts. This makes a lot of sense as well since they are solid performers for a good price, where HD2900 was not.

Are you sure the market data you're looking at isn't including IGP? I really have a hard time believing AMD hasnt gained major share in actual purchasers of mid range and up discrete cards. I really refuse to believe that. Now, maybe Nvidia has all the OEM contract for discrete cards locked down ( or whatever you would call cards contained in prebuilt systems), as far as that side goes. But there's no way I believe end user purchases of discrete cards haven't made major gains toward AMD this time around. No way. Now, I fully believe Nvidia has at least 50% of that market, but I'd wager in the R600 era they had 80%+ of it. I mean NOBODY was buying ATI cards back then.

Beyond that I dont know what to say, if people refuse to buy AMD cards simply because they are AMD, as you seem to suggest, well then it doesnt matter what AMD does so theres no point in them trying at all anyway, thus your strategy debate is irrelevant.

It is obvious Nvidia has put out a competitive product to AMD at every price point. Case in point, the GTX275 they introduced to compete with HD4890 is almost the exact same performance for the exact same price. Therefore if you like Nvidia cards, you almost always have an equal Nvidia option at any price point. Of course we all know the huge die size differences there which favor AMD.

Personally I figure the reason AMD doesnt cut prices more and really hurt Nvidia, as it seems they could, is simply that the CPU side is struggling so much they need every ounce of profit they can get get from the GPU side within reason. Realistically AMD CPU's dragging down the GPU's, it's kind of a nightmare for ATI.

Now, I guess if you're saying AMD needs to go for THE clear fastest card, period, then I'd have to agree with you on that. However, they are already pretty close with 4890. Theyre maybe 10% from GTX 285 now, once you OC a 4890 it's typically faster than a 285. And the dual GPU picture gets even murkier who is fastest. And supposedly theyre not even trying to be fastest, yet they compete with Nvidias top GPU as is. So again, I dont know what AMD has to do, but theyve clearly out-engineered Nvidia this go round (just as Nvidia had out engineered them in the last go round). I mean, it seems unlikely to me that AMD is ever going to me more than 10-20% faster than Nvidia, in general, as these companies are typically so competitive. And theyre within 10% now so..so basically if you're saying AMD needs to be 50% faster than Nvidia to start gaining major share, is that realistic?

It's important to note that Mercury includes both desktop and mobility in their discrete market share data.

 

[Sound_Card]

I admit having a split personality; meaning I'm answering my own questions:

Do I believe RV870 will be an outstanding D3D11 performance GPU?

Yes.

Is AMD going to retain for the next generation to the same degree their advantages when comparing RV7x0 to GT200?

No.

Is G92 outdated compared to GT200?

Both are D3D10 GPUs.

Personally I would have preferred AMD not to sell their handheld division and to concentrate even more aggressively on professional markets. IMHLO ATI has been limited way too much to something I'd call the "middle playground". I hope things will change in the less foreseeable future when this darned financial crysis is over and AMD could be able to divert consistently more and more resources to its graphics division in order for it to grow over the years.

Why don't you think RV870 will have a lesser degree of effect this next go around?
I could also tell you play "crysis" too much.

 

[Ailuros]

Why don't you think RV870 will have a lesser degree of effect this next go around?

Because I don't think AMD will have the same perf/mm2 advantage this round and yes its just a feeling and nothing else.

I could also tell you play "crysis" too much.

I have bothered it with it only a couple of times and never really finished it, so careful while jumping to conclusions.

 

[Sound_Card]

Because I don't think AMD will have the same perf/mm2 advantage this round and yes its just a feeling and nothing else.



I have bothered it with it only a couple of times and never really finished it, so careful while jumping to conclusions.

I was joking you. Crisis spelled crysis.

I'm expecting AMD to have even yet better performance per mm2 on RV870 due to 40nm and efficiency tweaks/re hauls. This is a given though, but what is not is perhaps density improved even more to what ever small extent.

 

[Ailuros]

I'm expecting AMD to have even yet better performance per mm2 on RV870 due to 40nm and efficiency tweaks/re hauls. This is a given though, but what is not is perhaps density improved even more to what ever small extent.

I don't expect G300 to be as big as everyone expects and I should remind you that both are manufacturing this time on 40nm and not 55 vs. 65nm.

 

[ninelven]

I'm expecting AMD to have even yet better performance per mm2 on RV870 due to 40nm and efficiency tweaks/re hauls.

I wouldn't expect too much in this regard... Of course faster GDDR5 will help... but both of these are larger transitions for Nvidia than ATI, so one would expect the performance gap in this regard to close.

 

[trinibwoy]

Trinibwoy, what data do you have that AMD lost share from the HD2900 days?

I would hope that share increased from one of their lowest points. Else it wouldn't be one of their lowest points right? The test here is not whether they increased share relative to when they had poor products but whether they've been able to capitalize on their perceived advantages with RV770.

Are you sure the market data you're looking at isn't including IGP? I really have a hard time believing AMD hasnt gained major share in actual purchasers of mid range and up discrete cards.

I don't think so. The latest Mercury report did break out the desktop discrete numbers showing a 5% gain for Nvidia in the last quarter. Now that could certainly be due to dumping written off inventory in the channel as SB suggested but we don't know.

Beyond that I dont know what to say, if people refuse to buy AMD cards simply because they are AMD, as you seem to suggest, well then it doesnt matter what AMD does so theres no point in them trying at all anyway, thus your strategy debate is irrelevant.

As Jawed pointed out you can't simply build a chip and call it a day. So I actually look at it as AMD not trying at all. Now that could be due to some underlying philosophy that brands marketing as evil and unnecessary or merely due to a lack of funds.

Personally I figure the reason AMD doesnt cut prices more and really hurt Nvidia, as it seems they could, is simply that the CPU side is struggling so much they need every ounce of profit they can get get from the GPU side within reason. Realistically AMD CPU's dragging down the GPU's, it's kind of a nightmare for ATI.

They made $1M last quarter, I don't think they have much room to do anything right now. Although the 40nm ramp might change that.

Now, I guess if you're saying AMD needs to go for THE clear fastest card, period, then I'd have to agree with you on that.

No I'm not saying that at all. What I'm saying is that this strategy is dangerous because it is based on the assumption that it will always result in a perf/mm^2 advantage. That assumption may hold in a situation where Nvidia is dedicating more transistor budget to general compute. I doubt that's going to change soon so I expect AMD to retain it's gaming perf/mm^2 crown. But can you honestly look at the big picture here and simply use gaming performance only as a measure of engineering proficiency? That approach assigns zero value to all of the innovations and hardware investments that made CUDA possible.

It's possible that AMD will catch up in time to catch the GPU compute wave but if they take too long they could end up in the same situation they're in with Quadro vs FireGL. JHH made a point about hardware being only one part of the product - AMD is far behind when it comes to software support. And the same thing is shaping up in the HPC world as well.

 

[CarstenS]

I don't expect G300 to be as big as everyone expects and I should remind you that both are manufacturing this time on 40nm and not 55 vs. 65nm.

G300 as in GT300 without Tesla?

 

[Sound_Card]

I don't expect G300 to be as big as everyone expects and I should remind you that both are manufacturing this time on 40nm and not 55 vs. 65nm.

I was going to bring that up in which point I agree. But if that is the extent in which the gap would be closed, then I'm far from impressed. Of course more needs to be done. RV770 vs GT200 on 55nm is still rather pathetic in terms of per/mm2.

 

[Ailuros]

I was going to bring that up in which point I agree. But if that is the extent in which the gap would be closed, then I'm far from impressed. Of course more needs to be done. RV770 vs GT200 on 55nm is still rather pathetic in terms of per/mm2.

No one can be sure of anything of course at this point, but I don't think the gap will close. My gut feeling just tells me that the difference won't be as big as it used to be. Look below someone already detected that the missing "T" wasn't necessarily a typo

G300 as in GT300 without Tesla?

 

[Jawed]

What I'm saying is that this strategy is dangerous because it is based on the assumption that it will always result in a perf/mm^2 advantage.

AMD's strategy is to not waste money building a GPU almost nobody buys. Instead use that money to refresh all products on a more frequent and more timely basis. And perhaps use that money in other places, too.

AMD has delivered 40nm. NVidia made noises about being first with 40nm supposedly:

http://vr-zone.com/articles/nvidia-prepares-for-40nm-gpus/6020.html?doc=6020

and so far shows no signs of producing anything. Regardless, AMD's spending shift, downwards, is arguably working. Though it's early days. RV870 has a hell of a lot to live up to.

That assumption may hold in a situation where Nvidia is dedicating more transistor budget to general compute. I doubt that's going to change soon so I expect AMD to retain it's gaming perf/mm^2 crown. But can you honestly look at the big picture here and simply use gaming performance only as a measure of engineering proficiency? That approach assigns zero value to all of the innovations and hardware investments that made CUDA possible.

The key problem for NVidia starting roughly now is that both OpenCL and D3D11-CS dilute CUDA.

So while NVidia theoretically will have more advanced GPUs if it pushes features on GT300 beyond D3D11/OpenCL (it seems pretty likely to me - though bearing in mind there's nothing about the chip design of GT200 that is more advanced for compute than RV770), NVidia has to persuade people that their platform is more suitable than either OpenCL or D3D11-CS to use that advantage directly.

And of course there's Larrabee which, assuming it arrives and is reasonably powerful, has such a "feature and flexibility advantage" over OpenCL/D3D11-CS/GT300, that Intel will prolly find itself fighting-off the hordes (there is that "Larrabee HPC? Non!" attitude to contend with within Intel, it seems). That is, of course, provided Intel has any tools that are worth a damn I've discovered recently that Ct is seen internally as an interim solution, so there doesn't seem to be much point in looking in that direction.

The problem for NVidia is that the roadmap never included competing with Intel/Larrabee. Scary stuff trying to put CUDA up against a chip that is barely anything other than a compute chip. NVidia's relying upon both mindshare and tool quality, so obviously Intel has to do more than just enter the ring.

It's possible that AMD will catch up in time to catch the GPU compute wave but if they take too long they could end up in the same situation they're in with Quadro vs FireGL. JHH made a point about hardware being only one part of the product - AMD is far behind when it comes to software support. And the same thing is shaping up in the HPC world as well.

We'll find out how serious AMD is with OpenCL's debut in the next month or two.

Jawed