After attending 6 different talks about GPUs from various researchers today, it's pretty clear that most the people I've talked to pretty much just use Cuda. T

After attending 6 different talks about GPUs from various researchers today, it's pretty clear that most the people I've talked to pretty much just use Cuda. The general consensus was that OpenCL has a lot of overhead from setup and less performance despite being more flexible, and nVidia is much more active in actually courting researchers, often handing out machines for free. Some of the speedups especially for quantum chemistry and other "embarrassingly parallel" computations is pretty nuts.