<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
</head>
<body>
<div dir="auto" style="direction: ltr; margin: 0; padding: 0; font-family: sans-serif; font-size: 11pt; color: black; ">
Also compile with Intel MKL. I have ran cases in a cluster with 64 cores using intel MKL and the default library in QE (netlib). I got 13 times better performance using intel MKL and near linear enhancement in speed with increasing number of cores.<br>
<br>
</div>
<div dir="auto" style="direction: ltr; margin: 0; padding: 0; font-family: sans-serif; font-size: 11pt; color: black; ">
Yasser Al Wahedi <br>
</div>
<div dir="auto" style="direction: ltr; margin: 0; padding: 0; font-family: sans-serif; font-size: 11pt; color: black; ">
Assistant Professor<br>
</div>
<div dir="auto" style="direction: ltr; margin: 0; padding: 0; font-family: sans-serif; font-size: 11pt; color: black; ">
Khalifa University <br>
<br>
</div>
<div dir="auto" style="direction: ltr; margin: 0; padding: 0; font-family: sans-serif; font-size: 11pt; color: black; ">
<div dir="auto" style="direction: ltr; margin: 0; padding: 0; font-family: sans-serif; font-size: 11pt; color: black; ">
Get <a href="https://aka.ms/ghei36">Outlook for Android</a></div>
<br>
</div>
<br>
<br>
<br>
<div class="gmail_quote">On Thu, Dec 7, 2017 at 6:14 PM +0400, "Lorenzo Paulatto"
<span dir="ltr"><<a href="mailto:paulatz@gmail.com" target="_blank">paulatz@gmail.com</a>></span> wrote:<br>
<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div dir="3D"ltr"">
<pre>On 07/12/17 12:40, Sitangshu Bhattacharya wrote:
> As per my experience, QE/VASP or any other such computations which uses
> openmpi supports 2^n core processors.
Just to be clear: QE can use any number of CPUs, being a power of 2 or
not does not matter.
kind regards
This is actually total number of
> core per physical processor multiplied by 2 (hyperthreaded), check by
> typing "htop" in your terminal. You may try locating the wall time
> results done at different no. of cores for your computations. The
> results might be a bit faster if you use 32 instead of 40 (and so on)
> and divide them into pools. Please note that inter-node communications
> in PWSCF is the main bottleneck and depends on the 100GBpS or more
> connectivity switch. Eagerly waiting for this to be taken care by QE
> experts!. Image parallelization is supported by the Ph.x code, however
> collecting the results is a bit tricky if you are doing 2nd or 3rd order
> corrections. Parallelizing diagonalization technique make my computation
> a bit slower, probably due to the the less no. of atoms involved or so.
> Check with -np 2 only, first on a simple structure like Si and then
> increase this no. to see if this is taking more/less time. You will get
> an idea atleast..
>
> Regards
>
> On Thu, Dec 7, 2017 at 2:50 PM, Amar Singh
> <amarsingh122014@rediffmail.com <mailto:amarsingh122014@rediffmail.com="">>
> wrote:
>
> Dear Dr. Bhattacharya,
> Thanks for the link. Though, being very naïve to such computation,
> I am not able to figure out the best possible set of parameters I
> should start with. Would be thankful if you can suggest some typical
> values for -np, -ni, -nk, -nt and -nd for the system with 40
> processors and 256 GB RAM.
> thanks
> Amar
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum@pwscf.org <mailto:pw_forum@pwscf.org>
> http://pwscf.org/mailman/listinfo/pw_forum
> <http: pwscf.orgmailmanlistinfopw_forum="">
>
>
>
>
> --
> **********************************************
> Sitangshu Bhattacharya (সিতাংশু ভট্টাচার্য), Ph.D
> Assistant Professor,
> Room No. 2221, CC-1,
> Nanoscale Electro-Thermal Laboratory,
> Department of Electrical and Communication Engineering,
> Indian Institute of Information Technology-Allahabad
> Uttar Pradesh 211 012
> India
> Telephone: 91-532-2922000 Extn.: 2131
> Web-page: http://profile.iiita.ac.in/sitangshu/
> Institute: http://www.iiita.ac.in/
>
>
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum@pwscf.org
> http://pwscf.org/mailman/listinfo/pw_forum
>
--
Lorenzo Paulatto - Paris
_______________________________________________
Pw_forum mailing list
Pw_forum@pwscf.org
http://pwscf.org/mailman/listinfo/pw_forum</http:></mailto:pw_forum@pwscf.org></amarsingh122014@rediffmail.com></pre>
</div>
</blockquote>
</div>
<img src="https://mail.pi.ac.ae/Images/ZayedYear.png">
</body>
</html>