Re: Parallelisation

[Rudolf Weeber]

Hi Ahmed Reza,
On Fri, Jan 14, 2022 at 02:01:32PM +0000, Ahmad Reza Motezakker wrote:
> Thank you for help. I tried it and below please find the information:
> 
> 
> mpirun -np 4 ./pypresso  .../maintainer/benchmarks/lb.py 
> --particles_per_core=20000 --lb_sites_per_particle 6
> 
> and I get average timing of :    1.321e-01 +/- 7.118e-04 (95% C.I.)  on 
> Intel® Xeon(R) Silver 4114 CPU @ 2.20GHz with 20 cores.
So that's some 70% slow than what I got. Depends on the hardware, i guess.
Make sure to deactivate as much as possible in myconfig.hpp. Especially 
disabling rotation (dipoles, virutal_sites_relative, engine) will do a lot for 
performance concerning the short range interactions,since disabling them 
reduces the memory footprint of the particles.
Also make sure you didn't compile with debugging (only relevant if you actively 
turned that on at some point).

Higher number densities of particles take longer, because there are more 
interactions between the particles, if they are close to each other.

And as I mentioned, if you can use the GPU LB that will mater a lot.

Regards, Rudolf