Parallel efficiency of Lammps simulation using 10000+ CPUs

Hi All,

Does anybody experience a benchmark work about Lammps simulation using CPU number up to 10000?

Now I plan to simulate a system comprising of 10^8 carbon atoms. The force field I used is a simple molecular mechanical potential which includes the bond, angle, dihedral angle and LJ Vdw potentials. I can obtain 10000+ CPUs for the computation. But I have no idea about the parallel efficiency to use such a large number of CPUs. If you have a similar experience, could you please give me some hints?



Please see: