The GT300/Fermi Thread

[Nedjo]

Because it will be based on a newer chip revision?

Again folks, Tesla is based on A2. Everyone worried about power consumption and SP counts because of HPC cards should only be worried if they are buying a Tesla card. There's a reason NVIDIA is waiting for A3 before releasing consumer GeForce cards.

well if A3 is slated for April-May how you plan on having GeForce Fermi in Q1?

[annihilat0r]

I don't get it, so Tesla GPU's will be based on A2 revision? So they have already ordered mass (well, sort of mass) production of A2 for Tesla, but meanwhile have redesigned A2 so now they're at A3?

[LinkinParkBoy]

Q1 = Tesla = CES (7 ~ 10 Jan)

[Bodkin]

Q1 = Tesla = CES

CES= CONSUMER electronic show, what use is telsa there?

[zalbard]

I'm surprised this has not been posted yet.

Fermi Fun Fact of the Week: GF100 graphics cards provide hardware support for GPU overvoltaging for extreme overclocking.

Sounds fun!

[highoctane]

I'm surprised this has not been posted yet.

Sounds fun!

It was posted in this thread earlier: http://www.xtremesystems.org/forums/...4&postcount=87

[zalbard]

It was posted in this thread earlier: http://www.xtremesystems.org/forums/...4&postcount=87

Weird, search didn't pick it up... I suppose it doesn't look inside quotes, *sigh*.

[Cybercat]

well if A3 is slated for April-May how you plan on having GeForce Fermi in Q1?

Even Charlie Degenerate thinks it'll be sooner than that.

[Solus Corvus]

295 is current/last gen, the same potential for driver optimizations will exist for the geforce gt300 as much as the 5870.

I agree. I'd even suggest that there would be MORE room for improvement in the fermi drivers because of it being a new arch and it will take them a while to iron everything out. We don't even know how good or bad the initial fermi drivers will be so there is a wide range of possibilities there. I just brought up driver optimizations because LeadHead keeps mentioning the old 295.

The point that was ignored though is that evergreen has a refresh on the horizon. We have no clue though about possible fermi refreshes yet. Fermi performing like a 295 would be a disappointment from my perspective because ATI could easily beat it with a strong enough refresh. That would put Nvidia in a bad position and I don't want to see that. I would prefer if both companies stayed relatively healthy.

We still do not know of how the architectural changes will affect geforce performance but its safe to say at minimum a doubling of performance of a single gt200.

Even if the final geforce shaders are 448 total thats still an over doubling of the 275/295 while having all the extra memory bandwidth. I would say its within reason to expect more performance than the 295 without the worry of sli profile optimizations.

I don't think it's safe to say that we can expect a doubling of performance. I'd suggest that double the performance of a singe GTX285 is the top of the probable range, not the bottom. As per Amdahl, 40% faster then evergreen, etc.

[Marios]

After GeForce Fermi gets out, ppl might want to upgrade their older 285GTX.
This way we might find used 285GTX in extremely low prices.
So in case fermi has lower performance than two 285GTX in SLI, ppl might not upgrade.
Actually this was Nvidia's policy by now.
Each new generation to have a bit better performance than the previous in SLI.
In case Fermi can't reach that goal we might consider it a failure.

[NaMcO]

GTX280 wasn't better than a 9800GX2 in all applications at launch, i had both back then. It became better only afterwards (drivers, etc). This doesn't mean Fermi will be better or worse than 2x240PS GT200's anyway, just stating a fact.

[Marios]

I compare 285GTX with 8800GTX actually. I see nothing in between worth mentioning as a new generation.
I expect the top GeForce Fermi model (single GPU) to have the performance of 2x285GTX in SLI to consider it a winner. Anything worse is bad for Nvidia.

[trinibwoy]

It will be great if that happens but to think that it has to beat an SLI setup to be successful is a bit silly. The vast majority of gamers/enthusiasts are not interested in multi-GPU setups for all the usual reasons. The HD 5870 is only 40-50% faster than last generation parts but that's not stopping people from buying them.

[annihilat0r]

Indeed. HD5870 is unable to beat last generation's Crossfire, or even the downclocked version of it (4870x2).

GTX 295 is 10% faster than HD5870 anyway. If Fermi can meaningfully beat GTX 295, it'll mean that there'll be a healthy difference between Fermi and HD5870 (more than the difference between HD5870 and HD5850) and that makes it good. Of course, what'll make it a success or a failure is the price point.

[ajaidev]

Did u all read the nvidia's facebook??

Heres what i found interesting:

NVIDIA
Oakridge did not cancel the project. The story was false. Check out www.legitreviews.com. They have the scoop there.
December 18 at 6:33am

NVIDIA
GF100 is shipping in Q1 folks. Patience is a virtue!
December 18 at 6:34am

[Solus Corvus]

It will be great if that happens but to think that it has to beat an SLI setup to be successful is a bit silly. The vast majority of gamers/enthusiasts are not interested in multi-GPU setups for all the usual reasons. The HD 5870 is only 40-50% faster than last generation parts but that's not stopping people from buying them.

I don't think it will matter. There are some people that will only buy Nvidia and will only look at a fermi as an upgrade to their 295. But most people won't care how it performs relative to the last generation, they will care how it performs compared to the current gen lineup from both companies. Either way they will have to compete on price/performance, naturally. But if performance isn't significantly faster then 5890/5990 then that means worse margins and more market erosion - not what NV needs right now.

Indeed. HD5870 is unable to beat last generation's Crossfire, or even the downclocked version of it (4870x2).

The 4870x2 wasn't downclocked, it was the same as the 4870.

[annihilat0r]

well, but not the same as the 4890, which was what I meant by "previous generation".

[Solus Corvus]

Well ok, but it's not really downclocked in the sense that the 295 or 5970 is. The 4890 was a refresh.

[Calmatory]

Indeed. HD5870 is unable to beat last generation's Crossfire, or even the downclocked version of it (4870x2).

CPU bound.

[annihilat0r]

CPU bound.

That's why there's virtually no difference in FPS between 3GHz and 4GHz at high enough resolutions, even with a 5970?

[Seu]

Watercool an SLI of Fermi GTX would'nt be easy

[trinibwoy]

Well ok, but it's not really downclocked in the sense that the 295 or 5970 is. The 4890 was a refresh.

People are making comparisons to GTX 285 SLI, hence 4890 is the corresponding comparison on the AMD side.

CPU bound.

Lol, what? The 5870 is obviously not CPU bound.

[Solus Corvus]

It's possible that the 5870 doesn't match 4890CF because they spent some of the extra power on IQ improvements like angle independent AF and more AA modes. Or the drivers aren't mature at this point. We will know after a few more driver versions.

Fermi will probably be competing with 5x90 and that could be just a higher binning or an actual refresh. To keep margins high with an expensive chip and expensive board they will want to have a significant performance lead. Otherwise this could be the generation that brings ATI and Nvidia market share into parity. I hope they can pull off 40% faster then 5870, but I'm keeping my expectations low.

People are making comparisons to GTX 285 SLI, hence 4890 is the corresponding comparison on the AMD side.

Right. But it doesn't really matter because Fermi won't be fighting GTX285 SLI or 4890CF.

[LedHed]

I just hope they beat out the 295 with a single chip, then I'll be getting Fermi in a heart beat.

[trinibwoy]

Right. But it doesn't really matter because Fermi won't be fighting GTX285 SLI or 4890CF.

Yep, its competition will be slower than that.