A lot of excitement and mystery has surrounded Nvidia’s Fermi. What are its specs going to be? Will it beat the reigning single core graphics performance king, the 5870? Well as of today some of that mystery has been revealed; Nvidia officially lifted the Non-Disclosure Agreement (NDA) on some of the details regarding the GF100 GPU. The GF100 has a 512 CUDA processor (unified shader cores), 16 geometry units, 4 raster units, 64 texture units, and 48 ROP engines. It will feature full support of DirectX 11 and the memory will be connected to the 384 bit GDDR5 memory bus. Fermi also has a totally new way of addressing memory that includes 768KB L2 cache that is shared between the four graphics processing clusters.
A more interesting piece of news; not only have some of the specifications been released, but the GF100 has also been benchmarked and it has no problem running Far Cry 2. The card ran Far Cry 2 at a resolution of 1920×1200 with 4x AA and all other settings at Ultra high and it got up to 84 frames per second. The reference card hovered around 50 frames per second with the same settings. And these cards (the Fermi and the reference) were run with an i7 960 CPU with 6gb of memory. Note that it was early rumored that the reference card was a 5870, however later rumors suggested that the reference card was a GTX 285 instead. The true identity of the reference card has not been confirmed.
This early benchmark suggests that the Fermi is doing a great job, but because this is only one test and the added confusion of what the reference card was, it is hard to tell the extent of its performance. So whether or not the card can reclaim Nvidia the performance crown has yet to be seen, and let us not forget that the elusive Fermi has yet to be seen as well.
Check out the video of the benchmark above.
Nvidia has a very suspicious track record in the last couple of YEARS when it comes to fairness and truth. They are always manipulating and distorting every last bit of information to suit their PR and marketing agenda. Was the driver used WHQL? Surely not. Was it tempered with to accentuate performance? Most probably. The reference card, why it was not disclosed? There is not a single good reasons that comes to mind. Was this reference card run into the same spec machine as the GF100? Doubtful. I will believe GF100 numbers coming only from website that have to sold their professional soul to Nvidia and there are not too many of them really.
The GF100 was an afterthought. The GPU features where “bolted on” a Tesla oriented architecture when the GT200 successor was scrapped. It only perform OK because of its super brute force approach. I don’t expect the Joules per frame to be stellar, far from it in fact.
Also, just a little correction. apparently Nvidia process yield are so catastrophic that a full 512 cores iteration is piratically impossible, save a infinitesimally low quantity of cherry picked die. The GF100 will have in fact only 448 cores, a 12.5% reduction. I’ll bet this Farcry 2 benchmark was run a overvolted, hypercooled 512 cores version that is NOT representative of what you will able to buy MASS IN QUANTITY.
It is my opinion that a refresh of the ATI 5970 (5975?) will end up cheaper, coller and faster that a castrated GF100 with 448 cores.
Ramon
Right Yakumo, its GTX285. So, no rumour – the man in video said himslef during the testing that the reference card is a GTX285. It also seems to be more or less consistent with the some tests done by Tom’s Hardware and AnandTech with gtx285 and MOST importanly ati5870. If GF100 pulls 80fps average at 4AA and 1920×1200, then it presumably beats the ATI5870 in Far Cry 2 though by 15-20fps? Let’s wait for other games, and real tests on real=marketed Ferms and see what it does and how much it costs…
Sorry, misheard it’s ‘gtx two 85′ said fast. nvidia GTX285.
he said “on the gx 25″ in the video, maybe he meant gts 250