CUDA SDK早就出來了(大概一年了), http://www.nvidia.com/object/cuda_home.html , 以前做 graphics-based General-Purpose computation on GPUs (GPGPU) 很麻煩, 必須用directX or openGL,還要把data encode 成 texture, 現在有 CUDA 真是很棒, 而且 CUDA exposes several hardware features that are not available via the graphics API.The most significant of these is shared memory, which is a small (currently 16KB per multiprocessor) area of on-chip memory which can be accessed in parallel by blocks of threads.

還有..
What kind of performance increase can I expect using GPU Computing over CPU-only code?

This depends on how well the problem maps onto the architecture. For data parallel applications, we have seen speedups anywhere from 10x to 200x.

壞消息是...CUDA到目前還是不支援VISTA!! 我還是用 graphics-based GPGPU 方式寫程式, 等到支援vista後又得改寫一遍

reference http://forums.nvidia.com/index.php?showtopic=36286