Error returned from CUDA function in CUDA-aware MPI multi-GPU test

Well, seems it was a module issue from the cluster I work on and the test passes now. I’m closing the issue then.