[QE-users] Parallelization nscf calculation

[Paolo Giannozzi]

1. Consider that a non-scf calculation has a computational cost of the 
same order as for a comparable scf one. If your k-points grid for the 
non-scf calculation contains N times more points than the scf 
calculation, expect that the former takes N times more than the latter.

2. Just use "mpirun -nk XX", with XX a divisor of the total number of 
processors (preferrably also a divisor of the total number of k-points), 
  such that 360/XX is smaller than the maximum FFT dimension along axis 
3. Don't use "-nb N" except for special cases

Paolo

On 4/27/23 15:19, Francesco Delodovici wrote:
> 
> Dear users,
> 
> I need to use wannier90 on top of quantum espresso to deal with a
> multilayer BiFeO3-LaFeO3 with 60 atoms overall.
> 
> I run the scf calculation to compute the Kohn-Sham orbitals on a 6x6x4 
> k-mesh,
> and then I need to refine the k-mesh to 18x18x10 with a nscf 
> calculation, asking for 220 bands.
> The calculation is spin-polarized to deal with antiferromagnetism in BFO 
> with
> the Hubbard corrections to the Fe d states included.
> 
> I employ the following parallelization parameters over 360 processors:
> -nk 36  -nb 10
> without OMP parallelization.
> 
> The problem here is that the nscf calculation does not complete within 
> 24 hours,
> so I wonder: can I change the parallelization options in order to obtain 
> a better scaling?
> What would be a better setting?
> I appreciate any suggestion, even though I know that parallelization 
> does not perform miracles.
> 
> Thank you,
> Francesco Delodovici, CentraleSupelec Paris-Saclay Université
> 
> _______________________________________________
> The Quantum ESPRESSO community stands by the Ukrainian
> people and expresses its concerns about the devastating
> effects that the Russian military offensive has on their
> country and on the free and peaceful scientific, cultural,
> and economic cooperation amongst peoples
> _______________________________________________
> Quantum ESPRESSO is supported by MaX (www.max-centre.eu)
> users mailing list users at lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/users

-- 
Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
Phone +39-0432-558216, fax +39-0432-558222