NVIDIA today has launched a new product line known as Tesla specially for the supercomputing market. According to TGDaily and Beyond3D, the Tesla C870 card is G80 based using Quado series based PCB but without display outputs.

The card is clocked at 575MHz core, has 128 shader processors at 1.35GHz with 1.5GB GDDR3 memories. NVIDIA claims that a single Telsa C870 card has a peak performance of 518GFlops that is able to match a combined performance of 40 x86 processors and consumes 170W of power. NVIDIA unveiled a D870 Deskside Supercomputer containing 2 Telsa cards hitting 1.036 TFlops consuming 550W power and S870 GPU Computing Server packing 4 Telsa cards hitting 2.072 TFlops consuming 800W if power. No SLI interface is required as one CPU thread will control a CUDA device so there is a linear increase in performance.

Telsa C870 card will be priced at $ 1499 and D870 at $ 7500 to be available in August while the Telsa server will cost some $ 12K to be available in Q4. A vendor even plans to sell a 12 TFLops Telsa server (24 Telsa cards) that will cost between $ 60-70K. Nvidia CEO Jen-Hsun Huang told TG Daily that Tesla will be strictly focused for the enterprise market and will not be making its way to the consumer market. “Perhaps in the future,” said Huang, “[this technology] could do physics on the PC, but this would need a Windows API.”

Telsa C870 Card

Telsa D870 Deskside Supercomputer

Telsa S870 GPU Computing Server