I understand that different MPI ranks can run concurrently on a single GPU. Can Hyper-Q support concurrent non-mpi applications?
Hyper-Q is part of the mechanism, but the software front-end that makes this possible is the CUDA MPS. It should be possible to use MPS to enable multiple processes to share a GPU whether MPI is being used or not.
A fully worked example of a non-MPI case is given here:
derived from the work here:
[url]https://devtalk.nvidia.com/default/topic/887822/fail-to-launch-cuda-mps/[/url]