For some reason, it looks like Intel MPI jobs running on Slurm don't work properly if you don't set
I_MPI_HYDRA_IFACE=eth0
The initial job on a cluster works, but all subsequent jobs fail. Still trying to track down the actual cause, but if we can set this in the impi module that would be great.
For some reason, it looks like Intel MPI jobs running on Slurm don't work properly if you don't set
I_MPI_HYDRA_IFACE=eth0
The initial job on a cluster works, but all subsequent jobs fail. Still trying to track down the actual cause, but if we can set this in the impi module that would be great.