[QE-users] parallelization of pw.x

[Pietro Delugas]

Dear Michal

3.7x with respect to what ?

your cut and paste refers to the wall time and the total cpu time per 
mpi task, they differ because you are using thread parallelism.

if you don't have memory issues I would try to increase the number of 
mpi processes decreasing the number of thread and usually when the 
number on MPI tasks is smaller than the dimesions of the fft grid it is 
better to avoid using nt.

Hope it helps

regards

Pietro

On 30/05/19 16:42, Michal Krompiec wrote:
> Hello,
> I am trying to run a calculation on a 2D slab with a bit of adsorbate 
> (119 atoms in total), and I would like to parallelize it as much as 
> possible. I am using a 3 3 1 Monkhorst-Pack grid (so I have 5 k-points).
> I tried using -npool 5 -nt 4 using 20 MPI processes and 5 threads per 
> process but, as it seems, the speedup was just 3.7x:
>    PWSCF        :   1d 4h27m CPU      7h43m WALL
> What could have gone wrong, is there anything "obvious" I can do to 
> diagnose the problem? I am using QE 6.4rc, compiled with gcc and 
> OpenMPI, without ELPA.
>
> Best regards,
>
> Michal Krompiec
>
> Merck KGaA and University of Southampton
>
> _______________________________________________
> 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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20190530/1a49c580/attachment.html>