CUDA. The GPU is the new Floating Point Coprocessor. (I think they are likely to be integrated on CPUs even for high performance use-cases, eventually. Althoug

CUDA. The GPU is the new Floating Point Coprocessor. (I think they are likely to be integrated on CPUs even for high performance use-cases, eventually. Although this is only happening very slowly...) It should be be programmed with vendor-neutral CPU instructions and if need be, trapped by the kernel and emulated or delegated appropriately. But all of this should be totally transparent to the user application.