[QE-users] Parallelizing QE/SternheimerGW Germanium

[grayj6]

So it looks like they implement two levels of parallelization; npool and 
nimage although nimage seems to be defined differently. Running on Np 
processors with Nk x Nk x Nk kpoints and Ni images, there must be a nice 
ratio of Ni to Nk and Np which is effecient.

"With the pool parallelization,
we distribute different k or q points to different processors. This 
method is very efficient for both G and W,
if the number of these points is a multiple of the number of CPUs. The 
image parallelization distributes G
vectors used in the linear response and Fourier transform to different 
CPU. For the calculation of W, this
parallelization is very efficient because the solution of
the linear-response equation for each G vector is independent of the 
other G vectors." from - Lambert, Henry, and Feliciano Giustino. "Ab 
initio Sternheimer-GW method for quasiparticle calculations using plane 
waves." Physical Review B 88.7 (2013): 075117.

On 2020-02-26 13:43, Paolo Giannozzi wrote:
> Parallelization levels do not appear spontaneously: they have to be
> implemented, sometimes with a lot of effort. Somewhere in the
> documentation of SternheimerGW it must be stated which parallelization
> levels are available: presumably, k-point and/or plane-wave
> parallelization. The two parallelization levels you mention are
> implemented only in a subset of QE codes.
> 
> Paolo
> 
> On Wed, Feb 26, 2020 at 7:31 PM grayj6 <grayj6 at rpi.edu> wrote:
> 
>> Hello,
>> 
>> I have a calculation of Germanium that I would like to run, but I
>> am
>> uncertain as to how to parallelize it. I have a kpt_grid of 4, 4, 4
>> and
>> a qpt_grid of 4, 4, 4. The number of bands, num_band is 18. I did
>> the
>> scf calculation on a kpoint grid of 8 8 8 0 0 0.
>> 
>> My current submit command is:
>> mpirun -np 4
>> /home/grayj6/data/q-e-qe-6.3/SternheimerGW/main/bin/gw.x
>> -npool 4 < gw.in [1] >gw.out
>> 
>> ... but I do not observe any noticeable speed ups when I add in
>> nbgrp or
>> ntask parallelization.
>> 
>> Thank you!
>> _______________________________________________
>> Quantum ESPRESSO is supported by MaX
>> (www.max-centre.eu/quantum-espresso [2])
>> users mailing list users at lists.quantum-espresso.org
>> https://lists.quantum-espresso.org/mailman/listinfo/users [3]
> 
> --
> 
> 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
> 
> 
> 
> Links:
> ------
> [1] http://gw.in
> [2] http://www.max-centre.eu/quantum-espresso
> [3] https://lists.quantum-espresso.org/mailman/listinfo/users
> 
> _______________________________________________
> Quantum ESPRESSO is supported by MaX 
> (www.max-centre.eu/quantum-espresso)
> users mailing list users at lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/users