Thanks for this suggestions. I am using ifort (version 12 update 3) and mkl. i am trying to make the .machine file <br>for k point parallel mode and inform you if any problem persists. <br><br><div class="gmail_quote">On Fri, Dec 30, 2011 at 5:32 PM, Laurence Marks <span dir="ltr"><<a href="mailto:L-marks@northwestern.edu">L-marks@northwestern.edu</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0pt 0pt 0pt 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">A comment on Peter''s comment. OMP_NUM_THREADS might have an effect,<br>
but the parameter is really MKL_NUM_THREADS if you are using Intel's<br>
mkl library -- and even then the compiler makes it's own decisions<br>
about how many it will use and this varies with version of the<br>
compiler (and probably what one has to set).<br>
<br>
Also...many of these things depend upon exactly what is in your<br>
computer, e.g. memory speed, bus speed as well as what else is<br>
running. Best is to do your own benchmarks, e.g. see<br>
<a href="http://www.wien2k.at/reg_user/benchmark/" target="_blank">http://www.wien2k.at/reg_user/benchmark/</a>. For instance, while on older<br>
computers Intel's hyperthreading was not good much to my surprise when<br>
I tried it (by accident) on a recent dual hex-core machine it was<br>
quite reasonable.<br>
<br>
N.B., the same holds for mpi -- very hardware and software dependent<br>
in my experience.<br>
<div class="HOEnZb"><div class="h5"><br>
On Fri, Dec 30, 2011 at 5:44 AM, Peter Blaha<br>
<<a href="mailto:pblaha@theochem.tuwien.ac.at">pblaha@theochem.tuwien.ac.at</a>> wrote:<br>
>> To "make" the file, you use an editor (nothing more or less) and putthe<br>
>> file in the directory where you are running the job. I have notlooked, but<br>
>> there is probably something in w2web to do this.<br>
>> N.B., the code does not use OMP anywhere (at present) so this will notdo<br>
>> anything.<br>
><br>
><br>
> Just a small correction: If you are using the mkl, OMP_NUM_THREADS will have<br>
> an<br>
> effect in the diagonalization part of lapw1 (which is often the main part<br>
> anyway).<br>
><br>
> However, while OMP_NUM_THREADS = 2 is quite effective and speeds up the<br>
> diagonalization<br>
> significantly, =4 is far less efficient and = 6 is usually useless.<br>
><br>
> PS: On a 6 core machine you should run ONLY 6 jobs at the same time (or 3<br>
> k-parallel<br>
> jobs with OMP_NUM_THREADS=2) !!!<br>
> Intels "hyperthreading", which mimics "12 processors" to you, should even be<br>
> switched<br>
> off (BIOS), because it will slow down the jobs.<br>
><br>
><br>
><br>
> --<br>
> -----------------------------------------<br>
> Peter Blaha<br>
> Inst. Materials Chemistry, TU Vienna<br>
> Getreidemarkt 9, A-1060 Vienna, Austria<br>
> Tel: +43-1-5880115671<br>
> Fax: +43-1-5880115698<br>
> email: <a href="mailto:pblaha@theochem.tuwien.ac.at">pblaha@theochem.tuwien.ac.at</a><br>
> -----------------------------------------<br>
> _______________________________________________<br>
> Wien mailing list<br>
> <a href="mailto:Wien@zeus.theochem.tuwien.ac.at">Wien@zeus.theochem.tuwien.ac.at</a><br>
> <a href="http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien" target="_blank">http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien</a><br>
<br>
<br>
<br>
</div></div><div class="im HOEnZb">--<br>
Professor Laurence Marks<br>
Department of Materials Science and Engineering<br>
Northwestern University<br>
<a href="http://www.numis.northwestern.edu" target="_blank">www.numis.northwestern.edu</a> 1-847-491-3996<br>
"Research is to see what everybody else has seen, and to think what<br>
nobody else has thought"<br>
Albert Szent-Gyorgi<br>
</div><div class="HOEnZb"><div class="h5">_______________________________________________<br>
Wien mailing list<br>
<a href="mailto:Wien@zeus.theochem.tuwien.ac.at">Wien@zeus.theochem.tuwien.ac.at</a><br>
<a href="http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien" target="_blank">http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien</a><br>
</div></div></blockquote></div><br>