Those of you who are tracking NVIDIA and GPGPU would be aware of CUDA 2.0 .. well the final version is released and is available on many platforms including Windows Vista. I had been experimenting a bit with CUDA APIs in my spare time on my Fedora 9 machine and 8500 GT card. Though I have a similar setup for my Vista machine, I didn’t take the pain of configuring it on Vista … it looked far simpler for me to use gcc on Fedora 9. I had to make some minor changes to the sample SDK code to compile with the new gcc compiler though.
So what have I done? I merely wrote a small program to add two arrays of limited size. This whole example is very simple demonstration of how to pass data from host CPU to the device (GPU), do the computation on GPU and get back the results. Well a lot of such examples are there in SDK samples .. but you see, if you don’t do it your self, you don’t get it ;-) The source is available here (deceptively called vecdot.tar.gz)… and what is this thing called documentation?