Nvidia Gets into SuperComputer Biz

Nvidia has introduced a GPU-based "personal supercomputer" that claims to deliver a cluster's worth of computing power at one percent of the cost.

Graphics developer Nvidia has been talking for years about harnessing the processing power of its GPU designs for purposes other than splattering alien entrails across gamers’ screens, and the company has finally unveiled its plans at the SuperCompute 2008 with the GPU-based Tesla Personal Supercomputer, which claims to deliver 250 times the processing power of a typical PC workstation…yet still maintain the workstation price tag. Although these systems aren’t going to suddenly find their way into gamers’ lairs, anybody who does serious work with massively parallel computing might give the Tesla design serious consideration—and Nvidia has already lined up Dell, Velocity Micro, Colfaz, AMAX, Penguin, Microway, Boxx, Western Scientific, and other partners to produce systems.

“We’ve all heard ‘desktop supercomputer’ claims in the past, but this time it’s for real,” said Microsoft Technical Fellow Burton Smith, in a release. “Nvidia and its partners will be delivering outstanding performance and broad applicability to the mainstream marketplace. Heterogeneous computing, where GPUs work in tandem with CPUs, is what makes such a breakthrough possible.”

The systems are built off Nvidia’s CUDA parallel computing architecture. The Tesla systems run either WIndows XP or Red Hat or SUSE Linux (64-bit recommended!), and feature 3 or 4 Tesla C1060 processors, each with 4 GB of memory and 240 scalar processor cores. The system CPU’s are either quad-core AMD Phenom or Opteron chips or quad-core Intel Core 2 or Xeon processors and push up to 102GB/s peak bandwidth per GPU. The systems’ offer an overall performance rating of 933 GFlops single precision and 78 GFlops double precision.

Units are available now at prices under $10,000.