Graphics processing units can do lots of computations in parallel. That is, in part, what makes them good at rendering omgxplosions and secksy laydees. However, architecturally, they're not really limited to rendering lovely graphics. There are other applications for such processing power.
You probably knew that already, however, because for some time now there have been people working on general purpose GPU problems. In fact, some usable stuff has already come out of the work. Folding@Home is available in a version that'll work on ATI X19xx GPUs. ATI's got its Stream Computing technology for GPGPU stuff, and we all knew it wouldn't be long before NVIDIA came along with an equivalent.
So here it is, NVIDIA's announced 'CUDA'. It stands for Compute Unified Device Architecture; lovely.
While ATI launched Stream on existing R580 hardware, either because the GPUs suited it or more likely were designed to some extent with it in mind, it seems NVIDIA's waited until G80 to introduce its GPGPU support. That said, at this stage of 'the game' the loss in time is of little consequence; it'll be at least a little while before we see GPUs doing different types of work in the mainstream.
Amongst NVIDIA's CUDA tools is a C compiler that'll turn your glorious C into code that's executable on supported GPUs. NVIDIA's touting CUDA as a means of accelerating a number of computationally intensive tasks used in industry and research, though we're sure the open saucers will find lots of lovely things to do with it too.
Get a bit more info on CUDA in NVIDIA's press release and if you still want more, hit the NVIDIA Developer's pages.
