[QE-users] mutlicore usage for GPU.

Paolo Giannozzi paolo.giannozzi at uniud.it
Sun Apr 28 11:55:20 CEST 2024


On 26/04/2024 23:00, Bhargab Kakati wrote:

> while using more 
> than one core (mpirun -np 8) the calculation becomes very slow but it is 
> way faster when I do "mpirun -np 1". Is there a reason for that?

of course there is a reason, and you mention it in the following line:

> I have only 1 GPU

You should run one MPI process per GPU. If you have 1 GPU, you don't 
even need MPI. You may find OpenMP useful if you have available CPU 
cores, but don't expect any serious speedup from OpenMP

Paolo
-- 
Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
Univ. Udine, via delle Scienze 206, 33100 Udine Italy, +39-0432-558216
AVAILABLE POSITION: 
https://physicslab.uniud.it/persone/paolo-giannozzi/prin2022


More information about the users mailing list