Fix memory problems by increasing the number of nodes

On Shaheen, the physical memory available for each compute node is 128 GB and the total number of CPU cores per node is 32. When we put 32 MPI tasks on each node, the memory available for each MPI task is 128 GB/32 = 4 GB. In some situations where the memory requirement per MPI task exceeds 4 GB and could not be further reduced by just increasing the number of MPI tasks, the jobs fail due to memory problems. To fix this issue, we can further increase the number of nodes (--nodes) while keeping the number of MPI tasks (--ntasks) unchanged. By doing this, the number of MPI tasks per node is reduced and the memory for each MPI task is increased (note: By default, the MPI tasks are evenly distributed over the allocated nodes).