Usage of NVTAGS for topology-aware GPU selection


Recently I have tried to use NVTAGS (Nvidia topology-aware GPU selection toolset dedicated to assigning GPUs to MPI processes for higher performance) to accelerate the benchmark of my application which happens to launch multiple MPI processes including both inter-node and intra-node GPUs.

I wonder if NVTAGS is able to detect the full topology of multi-node system and if not is there any way to resolve or walk-around this issue.

Thanks for your response in advance.