What kind of job do you run? Is it LCAO, Force field or PW? It depends on it. If you run the LCAO or PW, more mpi is better performance. (depending on the K-point) But if you run the force field, 1 node with multiple threads (ex, 8, 16) will be better performance. In fact, it depends on the hardware and conditions. You can test a few cases then you will be sure best one in your setting.