Nvidia has announced that it will make GPU boards for computing applications under the Tesla brand name. These boards will feature 128 parallel processors, providing up to 518 gigaflops of performance per board... so for example, if you had a high end motherboard with four PCIe 8x slots, you could have over two terraflops of math crunching power at your fingertips!

Nvidia will also apparently market a GPU computing server that will house up to eight Tesla GPU's in 1U of rack space - giving up to 4 terraflops per 1U case.

The Tesla will be supported by Nvidia's CUDA software development platform, which integrates a debugger/profiler, dedicated drivers and standard libraries, allowing the development of C applications capable of processing vast ammounts of data in parallel.

Cuda will be available for both Windows and Linux.