NV: Everything under control. 512-Fermi may appear someday. Yields aren't under 20%

[yh125d]

Perforrmance aside..............I look at the track record , It seems ATI is happy in second place.
It's almost like ATI and Nvidea planned it this way.

The usual cards in performance order and usually in price too.................>>

1. Nvidia} ultra HIGHEND 500$ but faster dual card
2. ATI } 450$ but 10% slower dual card

3.nvidea } high end 350$ but faster
4.ATI } 300$ 10% slower

5. nvidea and ATi about even in performance but Nvidiea 20$ more. (mid range)
6. lower end........do we really care.

This is the basic breakdown. Sometimes I realize there were exceptions.

Yep, that's about precisely how ATI intended it to be (except they likely expected GTX480 and 470 to perform higher relatively than they do, for 5870 to be about as fast as GTX470, just like last gen played out)

And this strategy is working very well so far for ATI, small die size at the cost of halo single GPU allows them to compete on price and multi-GPU scalability, and worry less about yields (smaller die = much better yields)

 

[edplayer]

In a nutshell thats what I was saying, but I just dont see a refreshed 58xx series scaling as good as a gtx series by looking at the overclocked benchmarks.

do you see that happening with TSMC's 40nm process?

I expect Nvidia's 28nm part to be a monster but when will it be available? My guess is sometime in 2011 but I wouldn't be surprised at all if it slipped to 2012.

and when you say refreshed 58xx, do you mean the same card at higher clocks or this SI rumor that has been going around?

 

[happy medium]

Yep, that's about precisely how ATI intended it to be (except they likely expected GTX480 and 470 to perform higher relatively than they do, for 5870 to be about as fast as GTX470, just like last gen played out)

And this strategy is working very well so far for ATI, small die size at the cost of halo single GPU allows them to compete on price and multi-GPU scalability, and worry less about yields (smaller die = much better yields)

And if ATI starts to make a better profit from this method I'm all for it.
Competition is good!

Didn't they just come out of the red money wise ?

 

[happy medium]

do you see that happening with TSMC's 40nm process?

I expect Nvidia's 28nm part to be a monster but when will it be available? My guess is sometime in 2011 but I wouldn't be surprised at all if it slipped to 2012.

and when you say refreshed 58xx, do you mean the same card at higher clocks or this SI rumor that has been going around?

I thought the next node was 28nm for both companies meaning the refresh comming to us Q4 or Q1 2011? Southern Islands and gtx 485 or whatever they are gonna call it.

If not, it's already pretty well known a higher clocked 58xx part will not cut it vs a respun gtx higher clocked part.

Edit:
Mabe some of the board partners can make non reference boards like this Palit gtx 470.

The dual-fan cooler is up to 12 degrees cooler and around 4 dB quieter when compared to the reference cooler solution.

http://www.fudzilla.com/content/view/18633/1/

 

[Genx87]

do you see that happening with TSMC's 40nm process?

I expect Nvidia's 28nm part to be a monster but when will it be available? My guess is sometime in 2011 but I wouldn't be surprised at all if it slipped to 2012.

and when you say refreshed 58xx, do you mean the same card at higher clocks or this SI rumor that has been going around?

It all depends when TMSC actually delivers the process and what Nvidia's plans are for that process. AFAIK AMD is also waiting on this smaller process for their next chip.

 

[yh125d]

And if ATI starts to make a better profit from this method I'm all for it.
Competition is good!

Didn't they just come out of the red money wise ?

I think so, I think Q1 '10 was in the black. GF was keeping them overall in the red for a couple quarters but now that it's been spun off it's safe to say the ATI division is profitable

 

[SlowSpyder]

Perforrmance aside..............I look at the track record , It seems ATI is happy in second place.
It's almost like ATI and Nvidea planned it this way.

The usual cards in performance order and usually in price too.................>>

1. Nvidia} ultra HIGHEND 500$ but faster dual card
2. ATI } 450$ but 10% slower dual card

3.nvidea } high end 350$ but faster
4.ATI } 300$ 10% slower

5. nvidea and ATi about even in performance but Nvidiea 20$ more. (mid range)
6. lower end........do we really care.

This is the basic breakdown. Sometimes I realize there were exceptions.

But you are ignoring the fact that AMD has had the fastest single GPU for the last 6 months. That AMD still has the fastest card. Nvidia has the fastest single GPU (if you can find it) but it has some issues in the eyes of many people.

With the previous generation, Nvidia launched the GTX280 in June of 08, AMD was able to launch the 4870x2 in August of 2008. Nvidia launched the GTX295 in January of 2009. AMD launched the 5870 in September of 09, Nvidia launched Fermi in April (well, end of March, but we all know that was a paper launch/announcement type launch... not an actual product launch) but AMD has had the 5970 out for a while...

What I'm trying to point out is that AMD doen't sit happily in second place. The lead changes often. Of course there are exceptions as AMD had nothing for Nvidia during the 8800/9800 days and right now Nvidia still cannot match AMD's performance with the 5970.

In the end Nvidia may certainly end up faster if they can manage to get a dual Fermi part out. But does that mean they 'won' or that AMD is in 'second place'? It just means Nvidia will have the faster part for the time being, who knows what will happen after that... if anything, recent history shows AMD will be back on top with another part in the not too distant future.

 

[yh125d]

I thought the next node was 28nm for both companies meaning the refresh comming to us Q4 or Q1 2011? Southern Islands and gtx 485 or whatever they are gonna call it.

If not, it's already pretty well known a higher clocked 58xx part will not cut it vs a respun gtx higher clocked part.

Currently from AMD we're expecting SI at 40nm, which is sort of a half generation stopgap, to make up for TSMC and GF cancelling the 32nm nodes on which SI was intended. It's gonna be more than just a refresh (GTX280 ->285, 4870 -> 4890) but less than a full generation. We're expecting to hae it in Q4 (hopefully in time for black friday and xmas both.


From NV, who knows. We really haven't heard anything regarding a refresh of sorts, cause this gen is still brand new to them. They only have two parts out atm, and the third isn't expected til June. They have a lot to do with expanding the chip stack before really worrying about a refresh. We may not necessarily see anything too significant til their next gen at 28nm, it's too early to tell, but methinks a 512sp GTX485, and maybe, some sort of dual GPU card. I wouldn't expect either til yields get much better and NV can respin for better thermals/power (think along the lines of the late C3 GTX260s we saw last gen when 55nm was very mature, which could OC from 576 to 800mHz even

 

[happy medium]

But you are ignoring the fact that AMD has had the fastest single GPU for the last 6 months. That AMD still has the fastest card. Nvidia has the fastest single GPU (if you can find it) but it has some issues in the eyes of many people.

With the previous generation, Nvidia launched the GTX280 in June of 08, AMD was able to launch the 4870x2 in August of 2008. Nvidia launched the GTX295 in January of 2009. AMD launched the 5870 in September of 09, Nvidia launched Fermi in April (well, end of March, but we all know that was a paper launch/announcement type launch... not an actual product launch) but AMD has had the 5970 out for a while...

What I'm trying to point out is that AMD doen't sit happily in second place. The lead changes often. Of course there are exceptions as AMD had nothing for Nvidia during the 8800/9800 days and right now Nvidia still cannot match AMD's performance with the 5970.

In the end Nvidia may certainly end up faster if they can manage to get a dual Fermi part out. But does that mean they 'won' or that AMD is in 'second place'? It just means Nvidia will have the faster part for the time being, who knows what will happen after that... if anything, recent history shows AMD will be back on top with another part in the not too distant future.

I agree ATi is not doing a good job playing catchup latley,and so far this round is doing a great job.

I think this is mostly due to Nvidea shooting themslves in the foot by aiming to high, just like with the fx5800 series.

Yes Nvidia did fumble the ball but usually they come back with a interseption for a long touchdown.
like that analogy?

I think once again,just like before, a 8800gtx monster will soon strike.

 

[edplayer]

I thought the next node was 28nm for both companies meaning the refresh comming to us Q4 or Q1 2011? Southern Islands and gtx 485 or whatever they are gonna call it.

28nm is next but there is nothing about it that mandates it comes out this winter.

SI rumors are for it to be on 40nm process. Haven't heard any rumors on next Fermi version. My expections are mid 2011 at the earliest for any "high-end" video card. And like I said earlier, wouldn't be the least bit surprised if it turned out to be 2012.

 

[Lonyo]

28nm is next but there is nothing about it that mandates it comes out this winter.

SI rumors are for it to be on 40nm process. Haven't heard any rumors on next Fermi version. My expections are mid 2011 at the earliest for any "high-end" video card. And like I said earlier, wouldn't be the least bit surprised if it turned out to be 2012.

Not only that, but the PROCESS is only scheduled for Winter as far as I can tell.
When you consider how long it was between 40nm availability (lets say HD4770) and 40nm high end GPU production, you're looking at a long time, which would mean Q2-Q3 2011 for a high end 28nm GPU.

So don't expect 28nm high end GPUs at a set time, just hope that they come as soon as possible.

 

[happy medium]

Currently from AMD we're expecting SI at 40nm, which is sort of a half generation stopgap, to make up for TSMC and GF cancelling the 32nm nodes on which SI was intended. It's gonna be more than just a refresh (GTX280 ->285, 4870 -> 4890) but less than a full generation. We're expecting to hae it in Q4 (hopefully in time for black friday and xmas both.


From NV, who knows. We really haven't heard anything regarding a refresh of sorts, cause this gen is still brand new to them. They only have two parts out atm, and the third isn't expected til June. They have a lot to do with expanding the chip stack before really worrying about a refresh. We may not necessarily see anything too significant til their next gen at 28nm, it's too early to tell, but methinks a 512sp GTX485, and maybe, some sort of dual GPU card. I wouldn't expect either til yields get much better and NV can respin for better thermals/power (think along the lines of the late C3 GTX260s we saw last gen when 55nm was very mature, which could OC from 576 to 800mHz even

So a Nvidia respin and a 28nm refresh for ATI? Q4?
Thanks for the clarification. Is this a guess or has it been written somewhere?

 

[yh125d]

I agree ATi is not doing a good job playing catchup latley,and so far this round is doing a great job.

I think this is mostly due to Nvidea shooting themslves in the foot by aiming to high, just like with the fx5800 series.

Yes Nvidia did fumble the ball but usually they come back with a interseption for a long touchdown.
like that analogy?

I think once again,just like before, a 8800gtx monster will soon strike.

A more fitting comparison than FX5800 would be fermi ~= 2900xt. Ambitious, big, hot, power hungry, but not too terribly bad on performance and priced well. Eventually led to RV770/870. NV will learn from fermi that they -have- to start worrying more about performance/power - or else their dual GPU card days as we know them will be over, and using smaller dies, so they don't get bit in the ass again by yield issues

 

[happy medium]

28nm is next but there is nothing about it that mandates it comes out this winter.

SI rumors are for it to be on 40nm process. Haven't heard any rumors on next Fermi version. My expections are mid 2011 at the earliest for any "high-end" video card. And like I said earlier, wouldn't be the least bit surprised if it turned out to be 2012.

So you think it will take 18 months to get a fermi card at 28nm?
I totally disagree.
I think more like 9 to 12 months at most.

 

[yh125d]

So a Nvidia respin and a 28nm refresh for ATI? Q4?
Thanks for the clarification. Is this a guess or has it been written somewhere?

No, a 40nm half-gen/refresh/whatever from ATI, who knows from NV. Really since NV is only entering the game 6 mos. later they don't necessarily need to release a refresh part before 28nm like ATI does


And I don't know how much of that info came directly from AMD, but it's the commonly accepted rumor as it stands

 

[happy medium]

A more fitting comparison than FX5800 would be fermi ~= 2900xt. Ambitious, big, hot, power hungry, but not too terribly bad on performance and priced well. Eventually led to RV770/870. NV will learn from fermi that they -have- to start worrying more about performance/power - or else their dual GPU card days as we know them will be over, and using smaller dies, so they don't get bit in the ass again by yield issues

Correct, the 2900xt to the 38xx series was nice. Good call.

I really see gpu makers going to dual core chips on the same die next generation. Like the cpu's of today.
No more of these huge single core monsters.

 

[happy medium]

I thought that ATi and Nvidea were going for new designs every 18 months with a half node refreshes somewhere in between? A.K.A gtx 260 65nm to 55nm. No die shrink for Fermi?

 

[SlowSpyder]

I agree ATi is not doing a good job playing catchup latley,and so far this round is doing a great job.

I think this is mostly due to Nvidea shooting themslves in the foot by aiming to high, just like with the fx5800 series.

Yes Nvidia did fumble the ball but usually they come back with a interseption for a long touchdown.
like that analogy?

I think once again,just like before, a 8800gtx monster will soon strike.

I have no doubt Nvidia will make great products in the future based off of Fermi's architecture.

I'm sure Nvidia will score a touchdown. Then AMD will score one. And it'll go back and forth for the most part, once in a while one team will fumble. That's what I see the 2900 as. What you can argue that Fermi is too.

Nvidia may very well release a part that dominates like the 8800GTX. Of course AMD could release something that is 9700Pro-like as well. So far, in the DX11 generation, I'd say AMD has dominated. One only has the lead for so long before the other seems to come storming back.

 

[BenSkywalker]

NV will learn from fermi that they -have- to start worrying more about performance/power - or else their dual GPU card days as we know them will be over, and using smaller dies, so they don't get bit in the ass again by yield issues

That works under the assumption that they will be bit in the ass by yield issues and also ignores performance/power in GPGPU tasks. I'm not saying how it will be played out, but is there a difference if nV sells 10Million parts to gamers total versus 2Million parts to gamers, 5Million parts to video editors and 4Million parts to insurance analysts? Their performance/power for certain aspects of Fermi is in absolute terms dominant. nV is looking long term with their architectural direction and looking at broader markets. Again, I'm saying how it will play out, just pointing out that they are using a clearly different approach then AMD and that it could very well end up being that both AMD and nV prove to be quite successful with their different directions.

 

[happy medium]

I have no doubt Nvidia will make great products in the future based off of Fermi's architecture.

I'm sure Nvidia will score a touchdown. Then AMD will score one. And it'll go back and forth for the most part, once in a while one team will fumble. That's what I see the 2900 as. What you can argue that Fermi is too.

Nvidia may very well release a part that dominates like the 8800GTX. Of course AMD could release something that is 9700Pro-like as well. So far, in the DX11 generation, I'd say AMD has dominated. One only has the lead for so long before the other seems to come storming back.

I see the fx5800 and 2900xt as epic fails, the gtx 470/80 are not. In fact if they weren't so hot and power hungry ,I'm sure we would not even be having this conversation. We would be playing games on our new gtx cards. :thumbsup:

 

[happy medium]

That works under the assumption that they will be bit in the ass by yield issues and also ignores performance/power in GPGPU tasks. I'm not saying how it will be played out, but is there a difference if nV sells 10Million parts to gamers total versus 2Million parts to gamers, 5Million parts to video editors and 4Million parts to insurance analysts? Their performance/power for certain aspects of Fermi is in absolute terms dominant. nV is looking long term with their architectural direction and looking at broader markets. Again, I'm saying how it will play out, just pointing out that they are using a clearly different approach then AMD and that it could very well end up being that both AMD and nV prove to be quite successful with their different directions.

So in short your saying Fermi makes money in other places other then us gamers?

Very good points Ben.

 

[yh125d]

I thought that ATi and Nvidea were going for new designs every 18 months with a half node refreshes somewhere in between? A.K.A gtx 260 65nm to 55nm. No die shrink for Fermi?

That really only works if a new generation comes out close to a node change. GT200 and RV770 came out within weeks of each other, but were on different nodes initially. For Fermi to have gotten a die shrink shortly after release like GT200 did would mean delaying fermi to just before 28nm came out, which would have been another year, so that wouldn't have worked out

 

[happy medium]

That really only works if a new generation comes out close to a node change. GT200 and RV770 came out within weeks of each other, but were on different nodes initially. For Fermi to have gotten a die shrink shortly after release like GT200 did would mean delaying fermi to just before 28nm came out, which would have been another year, so that wouldn't have worked out

I guess it would have worked out that way if they didn't scrap the 32nm process ha?

 

[SlowSpyder]

I see the fx5800 and 2900xt as epic fails, the gtx 470/80 are not. In fact if they weren't so hot and power hungry ,I'm sure we would not even be having this conversation. We would be playing games on our new gtx cards. :thumbsup:

The 2900 is similar to Fermi in many ways... it came out well after the competition so us enthusiasts had higher hopes for them than what materialized. Both are hotter, more power hungry, and louder than their closest competition. Both are/were slower than the competitions fastest part (of course not that black and white seeing as AMD's fastest is dual GPU, therefore some may not choose to use it).

There are many parallels...

Nvidia gets to claim 'fastest single GPU', AMD could only claim that they were around the top four or so. That's really only where I see a huge difference between the two.

 

[yh125d]

That works under the assumption that they will be bit in the ass by yield issues and also ignores performance/power in GPGPU tasks. I'm not saying how it will be played out, but is there a difference if nV sells 10Million parts to gamers total versus 2Million parts to gamers, 5Million parts to video editors and 4Million parts to insurance analysts? Their performance/power for certain aspects of Fermi is in absolute terms dominant. nV is looking long term with their architectural direction and looking at broader markets. Again, I'm saying how it will play out, just pointing out that they are using a clearly different approach then AMD and that it could very well end up being that both AMD and nV prove to be quite successful with their different directions.

They -will- keep getting bit in the ass with yield issues is they keep selling 500-600mm^2 dies. Maybe not as bad as this generation with TSMC's own problem adding to the issue, but they could compete on price much better if they could squeeze the same performance out of a smaller die

Using GPGPU PPW to legitimize the power draw of a gaming card is fatuous. If NV wants to become big in the GPGPU market that's fine, but they need to keep their GPGPU products and their gaming products separate instead of cramming a compute card and a gaming card into one die, creating a bloated chip that most of GTX480 buyers will never use fully

Why can't NV make a good gaming card AND a good compute card? They don't need to do both with one die. It winds up with most of GTX480 buyers having tons of compute power they don't need/want, and most quadro/tesla buyers with a bunch of gaming prowess they'd just assume replace with more compute power.


It's like trying to build a sedan with a pickup bed and 4WD. You end up making a vehicle with sacrifices all over, and (hopefully) sooner or later you'll realize you should just build two vehicles, a car and a truck