Back at CES, Nvidia decided to save some money and to brief the NDA people about its new chip codenamed Femi. This was the first time when they decided to show the GF100 codenamed card to the NDA public and they decided to discuss the depth of architecture.

The Geforce part of Fermi architecture is codenamed GF100 but we still prefer to simply call it GT300. This is a 40nm, heavily delayed chip that will end up with 512 shaders (Cuda cores), 16 geometry units which is fifteen more than ATI, 4 Raster units, 64 Texture units, 48 ROP Units and 384 bit GDDR5 memory support.

Of course the card is good in DirectX 11 and Tesselation, one of its key features. All in all, the card is and looks fast but its biggest downside is that it comes later, possibly March.

Nvidia knows that the card is late, they don’t want to talk about what it looks like a six month delay: This is only true if Nvidia really manages to launch the card in March, as we heard that it might happen.

Nvidia didn’t talk about the performance, TDP and the actual clocks, as these things are yet to be finalized but performance wise it was said many times that GF100 single chip Fermi will end up faster than Radeon HD 5870, no doubt about it.

Let’s hope for Nvidia’s sake that they can actually ship the cards in volume in March, but something tells us that even at the actual hard launch, the company will claim that the sales will end up great and that high demand will might easily result in cards being unavailable, despite the effort to make a massive launch. This is yet to be seen.