Hello, everyone.
I have been using Quantum Espresso to do some calculations, and I noticed something in the output of the process. I use normally QE in Windows10, but I installed Ubuntu via WSL to get some codes that are not available in Windows. So, when I use QE in Windows, the output gives:
...
Parallel version (MPI & OpenMP), running on 8 processor cores
Number of MPI processes: 1
Threads/MPI process: 8
...
But when I do it in Ubuntu, the output gives:
...
Parallel version (MPI), running on 1 processors
...
The result is that my calculations in Ubuntu are much slower than in Windows. Can anyone explain this difference?
Best regards,
Ricardo Tadeu
There are 3 possibilities here, for what I see.
1) how QE was compiled? Did you use a parallele compiler on Linux?
2) how QE is run? Do you need to invoke some mpirun to run? I would try mpirun -n 8 pw.x to see if this works.
3) what environment does WSL provide to the Linux machine?
I am admittedly not an expert of WSL. For the type of message you sent I would concentrate on 2 and 3 as it seems QE see itself as a parallel compilation.
I hope this helps,
Roberto
Roberto D'Agosta, I found a way to do exactly what I wanted in this website:
https://www.quantum-espresso.org/Doc/user_guide/node11.html
We usually compile Quantum-Espresso doing:
./configure
make all
But instead, if we do:
./configure –enable-openmp
make all
Then pw.x does 8 threads/MPI same as I get in Windows10. However at first, it didn't make my calculation faster, there may be something I could change in WSL but I don't want to risk, and I will try again to check if it didn't make any difference.
Thanks Roberto D'Agosta
You should be aware that there are two way to parallelism. One is MPI and the other is OpenMP. Do please search online for an explanation of the two methods, their advantages and drawbacks.
In Windows I believe you are using MPI, in Linux you have activated OpenMP. In particular OpenMP might underperform for some calculations so you might not see a big improvement.
Moreover, look in the manual on how to instruct PW.x to parallelise the calculation. That produces usually a large improvement.
I hope this helps,
Roberto
Hello Ricardo Tadeu Maia
You can use my comment (https://gitlab.com/QEF/q-e/-/issues/568#note_1345268964) to build quantum espresso
Thank you, Roberto D'Agosta
And thank you Vadim Sazonov , I am learning about Parallelism, but I am only using QE to educational purposes, learning about the packages and about properties I can obtain from DFT. And then I plan to apply it in some works of myself.
Until now I have been doing simple calculations in my personal notebook. I use WSL for some packages like thermo_pw and EPW, but also I have QuantumVitas in Windows10, and I just realized calculations in WSL were much slower than in W10, that's why I asked how to build QE to use OpenMP.
So, I am not doing anything risky with WSL and QE, but I read your comments and I wondered how you did benchmark for QE.
Thank you very much
Bests
Ricardo
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Materials Science
- Materials
- Inorganic Materials
- Metals
- Alloys
- Steel