Accelerating ReaxFF simulations

Dear lammps user
I am running reaxff simulations with openmpi with 20-35 -np processors with 3000-4000 atoms. Still I feel that simulations are rather slow. Is there any way to accelerate them like some accelerator package?

thanks

Rajesh

Kokkos has a relatively recent implementation of ReaxFF.

Dear Dr Fiorin,

Thank you for your response. I installed kokkos package and the built lammps with make mpi. Is it enough to use kokkos? If yes please let me know which command I should use to run the script. Currently I tried with mpirun -np 32 -ppn 4 lmp_mpi -k on t 4 -sf kk -in in.lj. But what ppn signifies here? It doesn’t taking the command with -ppn option. And when I remove -ppn it is working but I dont see any speed improvement. Any suggestion.

Thank you.

-ppn is an option of mpirun, not LAMMPS.

What is your current speed of calculations? You can try tabulating long-range interactions of ReaxFF, see http://lammps.sandia.gov/doc/pair_reaxc.html

Michal