9
$\begingroup$

I want to simulate the attached model for a Cu NP in Ar using my GPU, but I got erratic results. All looks well when I simulate it in my CPU with the command:

mpirun --use-hwthread-cpus lmp -in in

and, in my understanding, it should be the same input card for a GPU but executed with the command

lmp -sf gpu -pk gpu 1 -in in

However, the results look very different between both. What do you think about this?

Thanks a lot!

https://drive.google.com/file/d/1ozgBJKeGdV7sORw_MYhkGrGVl9C0B534/view?usp=sharing

enter image description here

$\endgroup$
3
  • $\begingroup$ It looks like this issue may have been resolved: matsci.org/t/lammps-input-card-in-a-nvidia-gpu/37400/6. If you could summarize what fixed the problem, it would make a good answer here as well. Also, in general, if you post a question to multiple forums, its best to include links to the other versions in the question. This avoids users putting effort into a problem that may already have been solved elsewhere and allows future users to find the answer more easily wherever it is posted. $\endgroup$
    – Tyberius
    Jul 14, 2021 at 14:10
  • $\begingroup$ Hi Tyberius, thanks for your message. Yes, it is the same question! sorry for not put the link immediately, I was a very busy week. $\endgroup$ Jul 16, 2021 at 23:07
  • $\begingroup$ No worries, just wanted to make sure other users were aware the question was resolved. $\endgroup$
    – Tyberius
    Jul 17, 2021 at 2:23

1 Answer 1

3
$\begingroup$

This was answered on the matsci LAMMPS forum by Axel Kohlmeyer.

To summarize, there is a known bug (still unresolved as of 7/28/2021) that leads to a corrupted neighbor list when using GPUs in combination with a hybrid pair style. Axel suggests using the command lmp -sf gpu -pk gpu 1 neigh no -in in as a workaround.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .