Different results when using same number of cores (32 cores)

Dear Lammps developers,

I know that the results will be different when using different numbers of cores/processors, which has been discussed before and also shown on the errors page. However, I found divergent results from two same runs of a restart job with 32 cores (same numbers of cores) on a HPC. Is this normal? or should I worry about this divergence? I am using the stable version - 11 Aug 2017. I have also tested it with only one core, the results are identical. The time-integrated method in the runs is only “fix nvt”, and for the kspace_styple I used pppm. Would this set up introduce round-off errors when using same multiple numbers of cores?

Thank you for your time.

Regards,
Pengyu

Dear Lammps developers,

I know that the results will be different when using different numbers of
cores/processors, which has been discussed before and also shown on the
errors page. However, I found divergent results from two same runs of a
restart job with 32 cores (same numbers of cores) on a HPC. Is this normal?
or should I worry about this divergence?

​it depends. you are not providing sufficiently detailed information to
make a specific statement. there are many settings in the input that have
an impact, some of which are more visible when running in parallel. but you
also don't say what degree of divergence you are seeing. ​also, it could
simply be, that you are not restarting correctly.

​axel.​

I am using the stable version - 11 Aug 2017. I have also tested it with