Hi all, I am building a cluster for GPU intensive application called Relion. The requirement is that the mpi occupies processes in round robin fashion across the nodes. I have tried using -map-by node parameter to achieve the same. It failed to implement round robin algorithm. I am running mpirun.mpich as the executable since my mpirun corresponds to open mpi which is working in round robin fashion but is not capable for multi threading. Hence using MPICH2. Can someone help me in figuring out how to submit a mpi job to schedule in rr fashion. Compute nodes: 64 vCPU, 8 GPU. Parameters: 24MPI, 8 Threads I am looking forward to schedule 8mpi processes on each node instead mpich2 is scheduling 24 on one node. Hope I made my question clear. Regards, Rajeshwar G.