and in case it wasn’t clear, in this case you use a single MPI launch that results in multiple LAMMPS partitions that can run different simulations (and you would specify the total number of coprocessors on the node that you want to use for all partitions with the intel package arguments).