I've been wanting to speed up our build process for some data that involves a pretty good fit for something like CUDA et al. the only problem being that we run a mixed gpu environment, some Ati some nVidia.
So it'd become a maintenance headache because I'd have to support CUDA+cpu only, or only half the team would be able to build all of the data :/
Role on OpenCL so I can have some fun playing with it too!
I have written some GPGPU stuff but just using vertex and fragment shaders in OpenGL. Getting it to work like that should be possible and portable enough to run on Ati/nVidia. Unfortunately as soon as you say "should" my boss won't give me work time to investigate.
Andy