I ran the dipole example (https://github.com/lammps/lammps/blob/master/examples/dipole/in.dipole) with about 8M atoms on Ryzen core (not multicore) for 50 steps and the run time was 0:02:25.
I also ran on a M2000 GPU and saw the runtime was 0:02:29.
lmp_mpi -in in.dipole lmp_mpi -sf gpu -in in.dipole
Is that normal? Is there anything behind that?