Fermi hardware might still be
two months away, but NVIDIA has done the sage thing and released some tantalizing numbers and architectural details to keep the fanboys chirping in the meantime. The GF100 will signal the end of tiresome rebadging and clock speed massaging, and early adopters will find 512 CUDA cores, 48 ROPs, and a 384-bit GDDR5 memory interface sprawled across
three billion transistors. Big changes are also afoot in how the card will do its work, with a reorganization toward a more parallel workflow leading to promises of up to eight times better geometry performance than on the GT200.
HardOCP reports that anti-aliasing results have improved "notably," while the video we've got stashed after the break for you shows the GF100 beating the
GTX 285 handily in a
Far Cry 2 benchmark. Still, the
PC Perspective crew expressed some apprehension about the massive die size and how it could impact yields given the still young 40nm production process -- a sentiment echoed by other publications who questioned whether NVIDIA would not have been better off trying for a less ambitious, more gaming-oriented board. We should all know that answer soon enough.
Read - AnandTech
Read - Hot Hardware
Read - PC Perspective
Read - HardOCP
Read - Tom's Hardware
posted January 18th 2010 1:15AM
Good read though if you keep alert and draw your own conclusions, but that he says having such a large die and poweruse means there's nowhere to scale to in the future seems very true for instance.
And it's nice to see a non paid-for analysis, that's getting damn rare, so many sites are basically 'embedded' reviewers, that can say one semi-critical thing in an article, to make it seem real, but only if they keep to the script the rest of the time.
I hope the high end single GPU GF100 (GTX385 ??) is priced at $450 or less. Any higher and they will have missed the mark if performance is indeed only 10-20% better than the 5870.
Well it does seem they focused on a new angle a lot, full C compliance makes them almost very very fast CPU's, so perhaps they have some sort of plan behind it all, something we don't know yet, perhaps an directx that runs on the GPU itself or something, because I can't imagine they didn't see all this during development, and i can't imagine they focus so much on processing power over graphics without some plan, apart from selling cards to researchers and professionals that is.
But I could be wrong.