I’ve recently gotten access to some K20s and I’m quite interested in MPS, the Multi-Process Service that implements the Hyper-Q/MPI magic in CUDA 5.5.
Now, I know that PGI (or, should I say, NVIDIA? :) ) 13.7 doesn’t support CUDA 5.5 as a target toolkit to build with since it was just released a few days ago, but I was wondering if I could use cuda5.0,cc35 on a machine with CUDA 5.5 and use MPS as well? Or do I need to wait for cuda5.5 as a target?
Of course, if I can get MPS set up here, I’ll probably try, but perhaps you might head me off at the pass and save some work if it’s just not doable yet.