NVIDIA launched within the show SC09 (Supercomputing 2009) by Portland a first series of cards based on the architecture Fermi. These products are part of the Tesla range, we find four references that can be split into two separate lines.

The first workstation, adopting a format of conventional graphics card (the C2050 and C2070) and the second as a rack for server (S2050 and S2070). The Tesla C2050 and S2050 will ship 3 GB of GDDR5 while the C2070 and S2070 have 6 GB recall that memory is linked to the GPU via a 384 bit interface. The Tesla S2050 and S2070 include up to 4 GPUs with 12 and 24 GB of RAM.

Consumption of a TESLA C2050/C2070 or format the nearest future GeForce, vary between 190 and 225 watts. These and even more those of the S20 * 0 series are high, higher than the current generation. NVIDIA announces performance calculations double precision (64 bits) between 520 and 630 GFlops.In comparison, an ATI Radeon HD 5870 is 544 GFlops in double precision. This suggests rather low frequencies, although they are not yet fixed. Between these values and consumption, one may wonder if NVIDIA would not worry the 40 nanometers. Recall that the new GPU includes no less than 3 billion transistors, and is therefore particularly difficult to produce.

The firm, however, the chameleon has a little time to refine its products; the new Tesla will not be available until April 2010. It's always the SC09 show that NVIDIA has just introduced the first GeForce based on Fermi. You can find a graphics card rather long, fully faired, with external power connector PCIe 6 and 8 pins. Connectors for 3-way SLI are present, and at least one DVI port. The GeForce was turned on its 512 cores the first DirectX 11 benchmark, Heaven Unigine society. Recall that the GeForce Fermi is expected before the Tesla in the first quarter of next year.