[QE-users] QE Performance on cluster and AMD

[Baer, Bradly]

Paolo,

I believe there is the capability of creating a local scratch folder that is deleted on the end of the job.  I will try that as a solution as I don't think I am yet brave enough to go under the hood of QE.

As far as the Ryzen performance, I used the same open source libraries for both systems (OpenMPI, OpenBLAS, FFTW3).  I don't know how to identify if I have a conflict between MPI and multi threading, but I will look into using Intel's MKL and see if that solves the performance issues.  I will also keep searching; the mail you referenced reminded me that there is likely more interest in the performance of epyc rather than ryzen.

Thanks for the advice.

-Brad

--------------------------------------------------------
Bradly Baer
Graduate Research Assistant, Walker Lab
Interdisciplinary Materials Science
Vanderbilt University


________________________________
From: users <users-bounces at lists.quantum-espresso.org> on behalf of Paolo Giannozzi <p.giannozzi at gmail.com>
Sent: Saturday, April 25, 2020 10:30 AM
To: Quantum ESPRESSO users Forum <users at lists.quantum-espresso.org>
Subject: Re: [QE-users] QE Performance on cluster and AMD

On Sat, Apr 25, 2020 at 1:34 AM Baer, Bradly <bradly.b.baer at vanderbilt.edu<mailto:bradly.b.baer at vanderbilt.edu>> wrote:

My submitted jobs for phonon calculations keep being canceled because QE is writing large files to the networked filesystem which is both very slow and under a fairly tight quota.

don't you have anything else to use for scratch files? in the worst-case scenario, a local /tmp or /scratch file system, even if not visible to all processors, may do the job (with some caveats)

  Files in the _ph0  directory, and appear to be creating one file per processor that I request for my job. Is there some way to have QE load these files in to RAM instead

there is some way, but it requires some work and some understanding of how the "buffer" machinery used in the self-consistent code works. If you feel brave enough to work with the latest development version and to perform some tests, I can explain what to do.

A second question: has anyone had success in using QE on AMD Ryzen processors?  I have a Ryzen system that is struggling to match the performance of an Intel system that is several years older. Both are running QE6.5 on Ubuntu.  From pw.x output, cegterg is 10x slower and cdiaghg is 100x slower on the AMD machine.   I had hoped that going from 4 cores to 16 cores would make it faster, not slower!

100x means that something VERY wrong is happening. Are you sure there isn't some conflict between MPI and multithreading? Please also see the following message, that contains an important piece of information:
https://www.mail-archive.com/users@lists.quantum-espresso.org/msg37148.html<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.mail-archive.com%2Fusers%40lists.quantum-espresso.org%2Fmsg37148.html&data=02%7C01%7Cbradly.b.baer%40vanderbilt.edu%7Ca98d116614954748e48f08d7e92da395%7Cba5a7f39e3be4ab3b45067fa80faecad%7C0%7C0%7C637234254548330786&sdata=Y%2FcpGdYuEIi2mb5yJBhvBFM4e1RxSTAkebnAZztzCuY%3D&reserved=0>

Paolo



Thanks,
Bradly Baer
--------------------------------------------------------
Bradly Baer
Graduate Research Assistant, Walker Lab
Interdisciplinary Materials Science
Vanderbilt University


_________________
Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso<https://nam04.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.max-centre.eu%2Fquantum-espresso&data=02%7C01%7Cbradly.b.baer%40vanderbilt.edu%7Ca98d116614954748e48f08d7e92da395%7Cba5a7f39e3be4ab3b45067fa80faecad%7C0%7C0%7C637234254548340782&sdata=gYD0LRnmRcyRE7ne4lowIVxzc5aNalQxdlI7RRvxFM8%3D&reserved=0>)
users mailing list users at lists.quantum-espresso.org<mailto:users at lists.quantum-espresso.org>
https://lists.quantum-espresso.org/mailman/listinfo/users<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.quantum-espresso.org%2Fmailman%2Flistinfo%2Fusers&data=02%7C01%7Cbradly.b.baer%40vanderbilt.edu%7Ca98d116614954748e48f08d7e92da395%7Cba5a7f39e3be4ab3b45067fa80faecad%7C0%7C0%7C637234254548340782&sdata=BlSAR6HXWIHijzqS1BZQ%2FRW%2BXPAmKUzJ14IvSA%2FbgBg%3D&reserved=0>


--
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

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20200427/a744f7e9/attachment.html>