Hi! I am learning to use Lammps and have a new question regarding reaxff
in parallel with Lammps:
I have used the reax potential implemented in Lammps to simulate
some systems containing several thousands of atoms. I would like to know if there is some problem with the parallelization of the reaxff part of the code... I am asking that because in the machines we have here (cudas), despite the jobs are working with my input, always there is only one core
running the reax calculation. Is there some special flag or modification
in the compilation in order to make it to be really parallel ? Other potentials (Airebo, Tersoff...) are working just fine in parallel with no
Thank you in advance.
If you're saying that you run other LAMMPS jobs
in parallel and the initial output and final timing
list P processors, but when you run with the ReaxFF
you only see 1 proc in that output, then I don't see
how that could happen. How do you know ReaxFF
is just running on one core?
One thing that caught my eye was this excerpt:
“…in the machines we have here (cudas)…”
Does this mean the user is trying to run the ReaxFF component on GPGPU hardware? Is ReaxFF ported to run on GPGPU yet?
no, it isn't - if you are trying to run ReaxFF on the GPU
there is no command in LAMMPS to do that.
That is the problem, Steve. Other jobs list in their output p processors,
but with reaxff it lists always one processor...
Steve Plimpton writes: