The official GT300/Fermi Thread

[RAMMAN]

there are 4 papers on nvidia's website.
pg 19 of "fermi looking beyond graphics"

that doesnt sound very specific to me...

[003]

At some point, if not now, features intended to
boost compute performance may compromise the chip’s competitive position as an affordable graphics
processor.

Whatever, as long as it's still faster than the 5870, it will be entrenched as the fastest gaming GPU and the fastest GPGPU.

[Calmatory]

Whatever, as long as it's still faster than the 5870, it will be entrenched as the fastest gaming GPU and the fastest GPGPU.

...until HD5890 arrrives, that is if ATI can work wonders. If Nvidia is being slow, and ATI being fast.. ATI could hold the H5890 in it's sleeve and pull it out just to make GF100 look ugly.

[LordEC911]

At some point, if not now, features intended to boost compute performance may compromise the chip’s competitive position as an affordable graphics processor. At that juncture, the architectures may have to diverge — especially if the professional market grows larger than the consumer market.

That doesn't sound like they are doing that now. They are just stating that it may happen in the future but we will see.
Edit- Thanks. I only knew about the one whitepaper.

[DilTech]

...until HD5890 arrrives, that is if ATI can work wonders. If Nvidia is being slow, and ATI being fast.. ATI could hold the H5890 in it's sleeve and pull it out just to make GF100 look ugly.

Somehow I don't think a 5890 is ATi's answer to fermi, I'd think they're more hopeful for the 5870x2 to handle that task than the 5890.

[Calmatory]

Somehow I don't think a 5890 is ATi's answer to fermi, I'd think they're more hopeful for the 5870x2 to handle that task than the 5890.

True. Though depends how much ATI can squeeze out from HD5890. Definitely the HD5870 X2 will be the real answer, which beats the single card offering from Nvidia. But business wise it might not be that great product due to high costs.

[DilTech]

We'll know for sure where everything falls into place when NVidia launches. We can guesstimate 5870x2 performance by looking at CF 5870 results, assuming ATi don't have to downclock at all for the 5870x2, which at present we do not know.

[003]

...until HD5890 arrrives, that is if ATI can work wonders. If Nvidia is being slow, and ATI being fast.. ATI could hold the H5890 in it's sleeve and pull it out just to make GF100 look ugly.

GT300 is not going to be faster than the 5870 by such a negligible amount that a 5890 would pull ahead. There will be a 5870X2, but then again there will also be a dual GPU GT300.

[tajoh111]

http://www.dailytech.com/article.aspx?newsid=16401

ORNL to Use NVIDIA Fermi to Build Next Gen Super Computer

NVIDIA announced its new Fermi architecture at its GPU Technology Conference recently. The new architecture was designed from the ground up to enable a new level of supercomputing using GPUs rather than CPUs. At the conference, Oak Ridge National Laboratory (ORNL) associate lab director for Computing and Computational Sciences, Jeff Nichols, announced that ORNL would be building a next generation supercomputer using the Fermi architecture.

The new supercomputer is expected to be ten times faster than today's fastest supercomputer. Nichols said that Fermi would enable substantial scientific breakthroughs that would have been impossible without the technology.

Looks like NV has customers already for Fermi.

[Solus Corvus]

GT300 is not going to be faster than the 5870 by such a negligible amount that a 5890 would pull ahead. There will be a 5870X2, but then again there will also be a dual GPU GT300.

We don't even know how fast the GT300 is, much less how fast a 5890 will be, and you are already calling this one for NV? Lol...ok.

[russian boy]

http://www.ixbt.com/news/all/index.shtml?12/46/02 Translated At the end of this article it says that card in Huang hands was fake

[jfromeo]

I'd kill for just a number from any bench leakage now

[LiquidReactor]

GT300 is not going to be faster than the 5870 by such a negligible amount that a 5890 would pull ahead. There will be a 5870X2, but then again there will also be a dual GPU GT300.

If Nvidia is already on planning on making a dual G300 die gpu solution this should indicate that a single g300 will not completely obliterate hd 5870 in real world gaming. I'm guessing average 15-20% with highs of 30% faster in some games.

Also 40nM and 550mm^2 die size?
G300 x2 dual die solution will either have to be dual pcb or a single pcb thats longer then even the new 5870x2. GTX295v2 (single pcb) was hard enough despite the die shrink going to 55nm (576mm^2 >>> ~452mm^2). How nvidia is going to put two 550mm^2 dies together to form a single gpu solution is not something I want to think about on a technical level.

Somehow I don't think a 5890 is ATi's answer to fermi, I'd think they're more hopeful for the 5870x2 to handle that task than the 5890.

5870 reference pcb is pretty much designed with a higher power draw (higher clocked) chip in mind.

[Chumbucket843]

If Nvidia is already on planning on making a dual G300 die gpu solution this should indicate that a single g300 will not completely obliterate hd 5870 in real world gaming. I'm guessing average 15-20% with highs of 30% faster in some games.

Also 40nM and 550mm^2 die size?
G300 x2 dual die solution will either have to be dual pcb or a single pcb thats longer then even the new 5870x2. GTX295v2 (single pcb) was hard enough despite the die shrink going to 55nm (576mm^2 >>> ~452mm^2). How nvidia is going to put two 550mm^2 dies together to form a single gpu solution is not something I want to think about on a technical level.

no. nvidia doesnt want to have their gx2 card out late like they did with th 295 so they will have the fastest graphics so 5870x2 doesnt steal their thunder. the die is big because their is a lot of double precision units. i wouldn't be laughing at 3 billion transistors on a single chip. thats not an easy thing to do. the reason the 295 was a dual pcb card is because the memory bus size. that probably wont happen this time because the bus is smaller.

[BeyondSciFi]

I think we are missing the point here, just as most people missed the point of the GTX 280 vs. the Radeon 4870 battle. We shouldn't really care too much about the configuration of GPU cores, as this is besides the point. But more on that later, now let us do a quick numerical recap if we may.

The GTX 280 has 933 GFLOPS of overall computing power in single precision.

The Radeon 4870 has 1200 GFLOPS of overall computing power in single precision.

So, going by GFLOPS along, the Radeon 4870 should be the faster card in pretty much all gaming situations save the ones specially coded for Nvidia architecture (and possible driver errors). As we have seen from the dozens of reviews of these cards so far, in the vast majority of games including synthetic gaming benchmarks, we see the situation is reversed with the GTX 280 being the most consistent winner. But how is this possible? The more or less obvious answer is that gaming performance is not solely based on raw GFLOPS output. So if the GFLOPS is the not the number we should be looking at, then what? Well, I'm sure at least some of my colleagues here might attest to looking at actual game benchmarks. This is perfectly fine if, and only if, you want to compare GPU power in JUST that specific situation and NOT taking those results as OVERALL levels of performance. The reason for this is simple; there are many different gaming engines out there. Given this multiplicity of choices, even an average of say 10 or 20 of the current most popular games out there would still not be a totally accurate representation of the differing levels of performance between the GPUs. As some of us might know, in statistics, the average of a set is an artificial number which may not represent the initial conditions of that set. Basically, there is too much individual variation to use a simple average of an arbitrarily chosen group of games to call that results absolute. Thankfully, there is a way of finding the differing (if different) output between GPUs which is not arbitrary or somewhat subjective. The way to do this is by comparing another group of numbers, namely, the texture and pixel fillrates. Here is how the GTX 280 and Radeon 4870 stack up:

The GTX 280 has a texture fillrate of 48.1 GT/s and a pixel fillrate of 19.2 GP/s.

The Radeon 4870 has a texture fillrate of 30.0 GT/s and a pixel fillrate of 12.0 GP/s.

As we can see, the GTX 280 has a significant output advantage over the Radeon 4870. This advantage seems to manifest itself as higher gaming performance in most game engines, and seemingly reflected in various benchmarks. It should be noted that how a certain output is produced is beside the point. The number of Shaders, TMUs, ROPs and the speed they operated at is JUST a way of getting their output; texture and pixel fillrates. Think of it in terms of internal combustion engines, just as how the number of cylinders, valves, displacement, RPM, etc., of an engine are just means of getting desired results, i.e. torque and horsepower. And to continue on the relevant note of next generation GPUs, lets try to compare the flagships of the GT300 (probably the GTX 380) and R800 (the now known Radeon 5870).

The Radeon 5870 has a texture fillrate of 68.0 GT/s and a pixel fillrate of 27.2 GP/s.

The GTX 380 has a texture fillrate of 83.2 GT/s and a pixel fillrate of 31.2 GP/s.
(If the rumored specs from Wikipedia are to be believed: http://en.wikipedia.org/wiki/Compari...rce_300_Series)

So, it seems the GTX 380 may be faster than the Radeon 5870 (with mature drivers, if not earlier). The GTX 380 (using the rumored specs) has 22% more texture fillrate and 14% more pixel fillrate. Given that the difference in performance is not too extreme, I suspect the GTX 380 will beat the Radeon 5870 in a majority of games and benchmarks but not necessarily all.

[MirageSys]

I think we are missing the point here, just as most people missed the point of the GTX 280 vs. the Radeon 4870 battle. We shouldn't really care too much about the configuration of GPU cores, as this is besides the point. But more on that later, now let us do a quick numerical recap if we may.

The GTX 280 has 933 GFLOPS of overall computing power in single precision.

The Radeon 4870 has 1200 GFLOPS of overall computing power in single precision.

So, going by GFLOPS along, the Radeon 4870 should be the faster card in pretty much all gaming situations save the ones specially coded for Nvidia architecture (and possible driver errors). As we have seen from the dozens of reviews of these cards so far, in the vast majority of games including synthetic gaming benchmarks, we see the situation is reversed with the GTX 280 being the most consistent winner. But how is this possible? The more or less obvious answer is that gaming performance is not solely based on raw GFLOPS output. So if the GFLOPS is the not the number we should be looking at, then what? Well, I'm sure at least some of my colleagues here might attest to looking at actual game benchmarks. This is perfectly fine if, and only if, you want to compare GPU power in JUST that specific situation and NOT taking those results as OVERALL levels of performance. The reason for this is simple; there are many different gaming engines out there. Given this multiplicity of choices, even an average of say 10 or 20 of the current most popular games out there would still not be a totally accurate representation of the differing levels of performance between the GPUs. As some of us might know, in statistics, the average of a set is an artificial number which may not represent the initial conditions of that set. Basically, there is too much individual variation to use a simple average of an arbitrarily chosen group of games to call that results absolute. Thankfully, there is a way of finding the differing (if different) output between GPUs which is not arbitrary or somewhat subjective. The way to do this is by comparing another group of numbers, namely, the texture and pixel fillrates. Here is how the GTX 280 and Radeon 4870 stack up:

The GTX 280 has a texture fillrate of 48.1 GT/s and a pixel fillrate of 19.2 GP/s.

The Radeon 4870 has a texture fillrate of 30.0 GT/s and a pixel fillrate of 12.0 GP/s.

As we can see, the GTX 280 has a significant output advantage over the Radeon 4870. This advantage seems to manifest itself as higher gaming performance in most game engines, and seemingly reflected in various benchmarks. It should be noted that how a certain output is produced is beside the point. The number of Shaders, TMUs, ROPs and the speed they operated at is JUST a way of getting their output; texture and pixel fillrates. Think of it in terms of internal combustion engines, just as how the number of cylinders, valves, displacement, RPM, etc., of an engine are just means of getting desired results, i.e. torque and horsepower. And to continue on the relevant note of next generation GPUs, lets try to compare the flagships of the GT300 (probably the GTX 380) and R800 (the now known Radeon 5870).

The Radeon 5870 has a texture fillrate of 68.0 GT/s and a pixel fillrate of 27.2 GP/s.

The GTX 380 has a texture fillrate of 83.2 GT/s and a pixel fillrate of 31.2 GP/s.
(If the rumored specs from Wikipedia are to be believed: http://en.wikipedia.org/wiki/Compari...rce_300_Series)

So, it seems the GTX 380 may be faster then the Radeon 5870 (with mature drivers, if not earlier). The GTX 380 (using the rumored specs) has 22% more texture fillrate and 14% more pixel fillrate. Given that the difference in performance is not too extreme, I suspect the GTX 380 will beat the Radeon 5870 in a majority of games and benchmarks but not necessarily all.

I would have to agree. Good job with the speculation and research. Looking at the rumored specs and comparing them percentage wise, in increases from both series(ATi4k and GTX2), the GTX380 should have a pretty handy lead (assuming few architectural changes). Unfortunately the architecture has changed so such an assumption will likely be inaccurate.

[Jamesrt2004]

" I suspect the GTX 380 will beat the Radeon 5870 in a majority of games and benchmarks but not necessarily all." .... that would suck for nvidia, they NEED To pull out by quite a bit if they trade blows even in 1 or 2 games.. then its in target for 5890... and assuming ati (having smaller dies) will be easier/cheaper to make, there going to be better price/perf again

tbh i dont think its nvidia's fault.. as much as tmsc's... im pretty sure if tmsc didnt heave yield problems nvidia would be better equipped right now. but same could be said for ATI? but meh

[Slovnaft]

tbh i dont think its nvidia's fault.. as much as tmsc's... im pretty sure if tmsc didnt heave yield problems nvidia would be better equipped right now. but same could be said for ATI? but meh

at this point it's clear that globalfoundries was the right move for AMD and almost proven as a general rule that market-leading semiconductor groups should control their own processes.

[Tim]

http://www.ixbt.com/news/all/index.shtml?12/46/02 Translated At the end of this article it says that card in Huang hands was fake

Yes it is a fake :

http://www.fudzilla.com/content/view/15798/1/

[malik22]

there wouldn be dual gt300 unless nvidia know the 5870x2 will be faster then the single gt300.

[Smartidiot89]

" I suspect the GTX 380 will beat the Radeon 5870 in a majority of games and benchmarks but not necessarily all." .... that would suck for nvidia, they NEED To pull out by quite a bit if they trade blows even in 1 or 2 games.. then its in target for 5890... and assuming ati (having smaller dies) will be easier/cheaper to make, there going to be better price/perf again

tbh i dont think its nvidia's fault.. as much as tmsc's... im pretty sure if tmsc didnt heave yield problems nvidia would be better equipped right now. but same could be said for ATI? but meh

I've thaught about AMD having smaller dies for their GPUs but I don't think it necessarily meens better yields...

While Nvidia have the uberhuge die, they have lower clock frequencies but looking at AMD we see the opposite, a small die and really high frequencies. I might actually go as far as saying the yielddifference between AMD and Nvidias topmodels at negligable?

[Vit^pr0n]

there wouldn be dual gt300 unless nvidia know the 5870x2 will be faster then the single gt300.

I think we pretty much know the 5870x2 will be faster just by specs alone.

Judging by revealed specs and BeyondSciFi's post, we can assume this is going to be a repeat of the previous gen performance ( GTX260 = 4870, GTX280 = 20-30% faster than 4870 )

Yes, there are arch changes in the GT300. However, most of it is for non gaming applications from what's been revealed. The gaming side of it is almost like the 5800 series: Double everything ( Though in the GTX380's case, not everything is doubled )

[003]

Though in the GTX380's case, not everything is doubled

But don't forget. RV870 has had the ROPs and TMUs doubled to 32 and 80, respectively.

GT200 already has 32 ROPs and 80 TMUs. GT300 will have 48 ROPs and 128 TMUs, and have SPs increased by 2.13x, and have SP efficiency increase in the process.

[Helloworld_98]

there wouldn be dual gt300 unless nvidia know the 5870x2 will be faster then the single gt300.

there won't be a dual GF100 anyway due to power usage, maybe dual GF100b but that's a long way off.

[003]

there won't be a dual GF100 anyway due to power usage, maybe dual GF100b but that's a long way off.

No, nvidia has already stated there will be. It most likely will be dual PCB. There's nothing wrong with that, the 7950GX2 and the original GTX295 were both dual PCB as well.

If power is an issue (which it shouldn't be), there is always the possibility of having three power connectors rather than two, or slightly scaling down the chip, as with the GTX295.