The GT300/Fermi Thread

[Solus Corvus]

Yep, its competition will be slower than that.

You honestly believe that 5x90 cards will be slower then GTX285 SLI and 4890 CF? Do you know something the rest of us don't?

[orangekiwii]

You honestly believe that 5x90 cards will be slower then GTX285 SLI and 4890 CF? Do you know something the rest of us don't?

And you honestly believe the 5890 is gonna be 20%-30% faster than a 5870?

[Solus Corvus]

No, I don't. 10%-20% is probably a more reasonable expectation.

[trinibwoy]

I think you have a very optimistic view on HD5870 performance. 10-20% isn't enough to catch GTX 285 SLI.

[LedHed]

I think ATI is more worried about beating out the 295 with their single chip refresh, not SLI 285s

[trinibwoy]

Why would AMD be concerned with an EOL part?

[LedHed]

Why would AMD be concerned with an EOL part?

295 is the only 200 series still in production (other than that weird eVGA 275 + G92) and they would be concerned because it beats out the 5870 in 80% of titles.

[trinibwoy]

No I mean by the time this fabled refresh hits Fermi will be on the market and GTX 295 will be long gone. I'm waiting to hear what this refresh will be based on though given that we're stuck on 40nm for all of 2010. It could be a new architecture but it remains to be seen whether it'll be more efficient than Cypress on the same node.

[LedHed]

well yea when the 5870 refresh hits the 295 will definitely be EOL on production. I wonder what happened to the ASUS Mars 295s; did one person buy like half of them or did they all melt and weren't able to RMA due to limited production?

[annihilat0r]

if we're stuck on 40nm for 2010 this means that there isn't going to be a 5990 like previously hinted

[Solus Corvus]

I think you have a very optimistic view on HD5870 performance. 10-20% isn't enough to catch GTX 285 SLI.

It's enough to catch a 295 that everyone keeps mentioning in this thread. GTX285 SLI is 59x0 territory.

Why would AMD be concerned with an EOL part?

GTX285 is EOL too.

No I mean by the time this fabled refresh hits Fermi will be on the market and GTX 295 will be long gone. I'm waiting to hear what this refresh will be based on though given that we're stuck on 40nm for all of 2010. It could be a new architecture but it remains to be seen whether it'll be more efficient than Cypress on the same node.

It would be on the same node with higher clocks and some arch tweaks - just like 4890.

[orangekiwii]

You honestly believe that 5x90 cards will be slower then GTX285 SLI and 4890 CF? Do you know something the rest of us don't?

No, I don't. 10%-20% is probably a more reasonable expectation.

I think you have a very optimistic view on HD5870 performance. 10-20% isn't enough to catch GTX 285 SLI.

According to this graph and your first quote, 20% is NOT enough to catch "GTX285 SLI and 4890 CF"

It's enough to catch a 295 that everyone keeps mentioning in this thread. GTX285 SLI is 59x0 territory.


GTX285 is EOL too.


It would be on the same node with higher clocks and some arch tweaks - just like 4890.

What I, and I think everyone else here understands, think you're trying to say is

5890 =? GTX295

5990? =? GTX285 SLI =? Fermi?

[LedHed]

5890 =? GTX295

5990? =? GTX285 SLI =? Fermi?

295 and 285 SLI are only 10-15% different in speed (since the 295 is 2x 275) unless you game at huge resolutions and the memory bottleneck on the 295 is exposed. However this only happens at above 1920x1200 with full IQ.

[Solus Corvus]

According to this graph and your first quote, 20% is NOT enough to catch "GTX285 SLI and 4890 CF"

That's right - according to that graph.

Crank it up to 2560x1600 and/or add some more AA plus rerun it on the drivers that are out at Fermi's release and I think you'd see a different picture.

You're just contradicting yourself. Up above you state the 5x90 will compete with gtx285 SLI, and there won't be a 5990 or a 5790 (that competes with gtx285 SLI), that leaves the 5890 which will NOT compete with it like you previously quoted.

You're just contradicting yourself.

That's not what I'm trying to say. In fact I'm not sure I'm even interpreting this mess of words correctly.

What I, and I think everyone else here understands, think you're trying to say is

5890 =? GTX295

5990? =? GTX285 SLI =? Fermi?

Roughly.

[jfromeo]

5990?

If they already hit the PCI-e 2.0 TDP limit with the 5970 and had to lower both mem and core frequencies to stay below 300W, the 5990 it's very unlikely to happen (unless they use 32nm for the 5890? which is even more improbable ).

[trinibwoy]

Didn't you know about the unwritten rule that you have to predict the most favorable outcome for AMD and the worst possible outcome for Nvidia? The 32nm 5990 will be with us right quick!

[Solus Corvus]

5990?

If they already hit the PCI-e 2.0 TDP limit with the 5970 and had to lower both mem and core frequencies to stay below 300W, the 5990 it's very unlikely to happen (unless they use 32nm for the 5890? which is even more improbable ).

I agree that it is becoming increasingly hard for both ATI and Nvidia to keep their dual GPU solutions within 300W. That doesn't necessarily mean that they won't make a >300W 5990 if NV threatens their market position with Fermi.

32nm is too far off.

Didn't you know about the unwritten rule that you have to predict the most favorable outcome for AMD and the worst possible outcome for Nvidia? The 32nm 5990 will be with us right quick!

Instead of responding with sarcasm you could simply propose and argue for a more likely scenario.

[Calmatory]

32nm will not happen. 28 nm will be H2/2010. Probably Q4. One year away.

[god_43]

IT LOOKS LIKE we were right about Fermi being too big, too hot, and too late, Nvidia just castrated it to 448SPs. Even at that, it is a 225 Watt part, slipping into the future.

Update: Nvidia has contacted us and declined to respond.

The main point is from an Nvidia PDF first found here. On page 6, there are some interesting specs, 448 stream processors (SPs), not 512, 1.40GHz, slower than the G200's 1.476GHz, and the big 6GB GDDR5 variant is delayed until 2H 2010. To be charitable, the last one isn't Nvidia's fault, it needs 64x32GDDR5 to make it work, and that isn't coming until 2H 2010 now.

http://www.semiaccurate.com/2009/12/...-fermi-448sps/


eee if this is true....it does not bode well.

[LedHed]

I thought that site was going down? It spurs utter BS on a daily basis.

[trinibwoy]

Instead of responding with sarcasm you could simply propose and argue for a more likely scenario.

What would be the point of that? Didn't you just dismiss the hard numbers posted a little earlier and went on to predict great driver improvements for Cypress and crap drivers for Fermi? Seems you have it all sorted already.

In any case more bad news out of TSMC for Nvidia - not enough A3's are hitting the speeds needed for C2070, i.e 1400Mhz.

http://forum.beyond3d.com/showpost.p...postcount=2442

[Solus Corvus]

What would be the point of that? Didn't you just dismiss the hard numbers posted a little earlier and went on to predict great driver improvements for Cypress and crap drivers for Fermi? Seems you have it all sorted already.

What would be the point of discussing it instead of sarcastically beating up strawmen? To explore the topic from a different point of view perhaps? Or do you expect me to play devil's advocate against my own perspective as well?

I dismissed those numbers and I gave my reason why. I can elucidate further if necessary. I'm not even sure why you brought up GTX285 SLI, everyone else was talking about GTX295. And both of those cards are EOL as well as 4890.

You want my optimistic perspective? The most (realistically) optimistic performance figures I have heard for Fermi are 40% faster then evergreen. Let's say that ATI doesn't refresh the 5xxx series either so they have to compete with what they have. If that happens performance would probably, IMO, look something like: 5850 < 5870 ~= GTX360 < GTX380 < 5970 < GTX390
That's basically a redux of last generation. As such it would probably mean some more market share erosion and narrow margins for Nvidia, just like last gen. That's my optimistic view. If you wanted me to say NV would be king of the universe and there would be global peace I'm sorry, I just don't think it's going to happen.

[Iconyu]

I thought that site was going down? It spurs utter BS on a daily basis.

Actually it's spot on: http://www.nvidia.com/docs/IO/43395/...83-001_v01.pdf Unless you doubt the source...

[annihilat0r]

What would be the point of discussing it instead of sarcastically beating up strawmen? To explore the topic from a different point of view perhaps? Or do you expect me to play devil's advocate against my own perspective as well?

I dismissed those numbers and I gave my reason why. I can elucidate further if necessary. I'm not even sure why you brought up GTX285 SLI, everyone else was talking about GTX295. And both of those cards are EOL as well as 4890.

You want my optimistic perspective? The most (realistically) optimistic performance figures I have heard for Fermi are 40% faster then evergreen. Let's say that ATI doesn't refresh the 5xxx series either so they have to compete with what they have. If that happens performance would probably, IMO, look something like: 5850 < 5870 ~= GTX360 < GTX380 < 5970 < GTX390
That's basically a redux of last generation. As such it would probably mean some more market share erosion and narrow margins for Nvidia, just like last gen. That's my optimistic view. If you wanted me to say NV would be king of the universe and there would be global peace I'm sorry, I just don't think it's going to happen.

Except, if Fermi is indeed 40% faster than Cypress (which I don't believe is possible) GTX 360 won't be equal to 5870 but will be a lot faster than it, and GTX 380 will be extremely close to 5970, so price points being the same comparatively to the last gen (as in difference between GTX 275/285 and HD4800's), Nvidia will be in an über condition.

[Marios]

It seems Fermi's GP GPU is the production of Nvidia's chasing their Larrabee nightmares.
I hope they didn't really get so much hallucinated to forget the gaming part.