Calculation parallelization

Total energy, geometry optimization, DFT+U, spin....

Moderator: bguster

Locked
Vladyslav
Posts: 1
Joined: Sun Mar 08, 2020 4:11 pm

Calculation parallelization

Post by Vladyslav » Sun Mar 08, 2020 4:29 pm

Dear All,

I am new to Abinit and I am looking for a possibility to properly parallelize my calculations.
I have 10 cpu cores and use mpirun to parallel it.

My parameters:

Code: Select all

nband 200
ngkpt 8 8 8
paral_kgb 1
npkpt 1
npfft 1
npband 10
nspinor 1 nsppol 2  nspden 2
When I try to swap values for npkpt with values for npband I see 30 processes instead of 10 (using htop).
However, I have read in similar topics of this forum that npkpt paralleling has been recognised as the most effective.
I would like to know, whether my approach is correct, and if not - how to optimize it?

Thank you,
Vladyslav

ebousquet
Posts: 469
Joined: Tue Apr 19, 2011 11:13 am
Location: University of Liege, Belgium

Re: Calculation parallelization

Post by ebousquet » Mon Apr 20, 2020 8:41 am

Dear Vladyslav,
Abinit is parallelized over k-point by default such that you don't have to initialized the parallel_kgb variables. The only thing is that to be optimal the total number of k-points should be a factor of the number of CPU.
Otherwise, with paral_kgb you have to make sure that npband*npkpt is equal to the number of CPU.
Best wishes,
Eric

Locked