Nvidia Unveils Supercomputer DesignNvidia Unveils Supercomputer Design

Nvidia on Tuesday launched a high-performance computing design for workstations that would use the company's Tesla graphics processors for scientific computing.

Called the Personal Supercomputer, the design uses up to four Tesla C1060 GPUs, each with 240 processing cores. Each of the graphics cards slips into a PCI Express slot.

The C1060 GPU, introduced in June, offers 1 teraflop, or a trillion computational operations a second, and 4 GB of memory. Nvidia claims the GPUs deliver 10 times the computational power of a computer with two quad-core CPUs while using about the same amount of energy.

A workstation with four of the GPUs would deliver enough horsepower to handle large computational workloads common in scientific research and the oil and gas and medical industries. In order to run high-performance computing applications on the systems, however, the software's heavily computational portions would have to be run through Nvidia's CUDA compiler. CUDA, or compute unified device architecture, is a development environment based on the C programming language.

System builders and resellers offering Tesla-based Personal Supercomputers include Amax, Armari, Asustek, Azken Muga, Boxx, CAD2, CADNetwork, Carri, Colfax, Comptronic, Concordia, Connoisseur, Dell, Dospara, E-Quattro, JRTI, Lenovo, Littlebit, Meijin, Microway, Sprinx, Sysgen, Transtec, Tycrid, Unitcom, Ustar, Viglen, and Western Scientific. Starting prices for the systems are less than $10,000, according to Nvidia.

Nvidia and rival Advanced Micro Devices are aggressively pushing the use of their GPUs for high-performance computing. This year, Nvidia saw its technology being used for the first in one of the world's top 500 supercomputers. Ranked No. 29, the TSUBAME Grid Cluster at the Tokyo Institute of Technology also uses Intel Xeon and AMD Opteron chips and is built with NEC and Sun Microsystems hardware.