I asked some folks here and they said that you should look at using Pyxis which is a SLURM plugin allowing unprivileged cluster users to run containerized tasks through the srun command. It what we use on NVIDIA’s Selene system and would allow for the NVHPC SDK container to work as-is with pmix.
And if in another way, via a building a custom configured openmpi? With ompi_info we seen that NVIDIA HPC SDK Open MPI package was configured with --with-slurm but without --with-pmi:
Therefore, I cannot run my mpi program with SLURM via an interactive command srun.
I need to add only --with-pmi[x] and to recompile another Open MPI. Can I do this? If so, how do I specify the correct paths for it within an installed HPC SDK package?: $ ompi_info | grep pmi
`MCA pmix: isolated (MCA v2.1.0, API v2.0.0, Component v4.1.2)
MCA pmix: flux (MCA v2.1.0, API v2.0.0, Component v4.1.2)
MCA pmix: pmix3x (MCA v2.1.0, API v2.0.0, Component v4.1.2)
MCA ess: pmi (MCA v2.1.0, API v3.0.0, Component v4.1.2)`