Hello everyone, I'm using the latest version of LIGGGHTS and atom style superquadric on a university cluster. I want to do parallel computing of superquadric but it seems like that there is no difference in computational time for using 4 or 8 or 12 cores of CPU at all.
I wanted to ask if there is some rule for distributing CPUs for a simulation or some kind of lack of mpi library for parallel computing of superquadric atom_style?
I appreciate any helpand tips regarding this issue.
Thanks in advance, Hamed
mschramm | Mon, 03/08/2021 - 17:26
Number of atoms
Hello,
How many atoms per core are you running? I believe LIGGGHTS was recommending at least 10,000 atoms per core to get good scaling. With the superqudrics, you can probably use fewer due to how the contacts are calculated.
hamedhoori | Wed, 03/10/2021 - 21:04
Hi, dear mschramm,
Hi, dear mschramm,
Thank you for your reply.
Actually, I have multiple cases with a different number of atoms but mostly the number is between 8,000 - 20,000. I tested the case with different numbers of processors but for more than 4 processors the computation time won't change.