@acl, I've used it on Windows too. With a GTX550Ti card at home and a Quadro 4000 at work. It was a bit of a PITA to get it working at first, but well worth it as I got more than an order of magnitude speed up on some code that uses FFTs heavily. I do however get the occassional driver crash on the work system which I have not been able to debug.
In my experience once you have the examples working and have successfully compiled and run a simple CUDAFunction, you should be okay to go for it.