nVidia has just announced the beta release of the NVIDIA CUDA Software Developer Kit (SDK) and C-compiler for computing on NVIDIA graphics processing units (GPUs). This release is targeted at Windows XP and Redhat and the GeForce 8800 graphics card and even includes FFT and BLAS reference implementations.
The aim of the SDK is to let programmers get at the stream-y power of the GPU in C without having to first express their algorithms as graphics code. There are lots of types and directives to pile into the code to make it work, though, and it’s all got to be compiled with their purpose-built C compiler.
More info, with examples, at the CUDA homepage.