<div>>I am using LG Quad core processor machine with PWscf running on
linux. I want to know >whether there is any command to run it parallely
on the four processors or by default with >simple installation (i.e.,
./configure and make all) it uses all the four processors?</div>
<div> </div>It depends. configure detects your environment and set the options for an optimal "compilation".<br><br>If you have MPI well installed, then configure detects it , and you run it as Lorenzo explained (mpirun -np path-to-executables/pw.x )<br>
<br>There is another paralellism that you can exploit, although it is not from espresso, but from the Intel MKL implementation of the BLAS/LAPACK libraries. With ifort compiler (or other OpenMP capable compiler) and the Intel MKL libraries well installed, configure detects the MKL and sets the compilation options automatically to use the MKL. The MKL have OpenMP parallelism implemented, and you run pw.x like a serial program, but the matrix operations made by MKL libraries will run parallel. If it works, with the command top you will see <br>
CPU use higher than 100%, near 400 % if you have a QuadCore. <br><br>Take it into account if you compare performance of mpi vs "serial". When I was unaware of the MKL built-in parallelism got unhappy because mpirun -np 2 pw.x (in a Dual Xeon) was only 30 % faster than just pw.x <br>
<br>Generally, this is slower than paralellization with MPI, but there is at least one example where OpenMP runs faster. <br><a href="http://www.democritos.it/pipermail/pw_forum/2008-May/008870.html" target="_blank">http://www.democritos.it/pipermail/pw_forum/2008-May/008870.html</a><br>
and<br><a href="http://www.democritos.it/pipermail/pw_forum/2008-May/008884.html" target="_blank">http://www.democritos.it/pipermail/pw_forum/2008-May/008884.html</a><br><br><br>If you have Intel or AMD CPU, try to compile espresso with the MKL. For large calculations, the one for which speed is really important, using the MKL can provide a factor of 5 in speed. <br>
<br>In summary,<br>If you just want to do fast calculations for small systems, do not care of installing the parallel environment.<br>If you want to calculate large systems, take your time to study the manual. After some training one can set the system to run efficiently in half a day, i.e., install Fortran compiler, MKL libraries, MPI, ESPRESSO an do some tests. Note that the latests distributions of Intel Fortran Compiler include the MKl, then it is one step less. <br>
<br>Regards<br>Eduardo<br><br><div class="gmail_quote">
<blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">---------- Mensaje reenviado ----------<br>From: Neel Singh <<a href="mailto:neelphysics@yahoo.in" target="_blank">neelphysics@yahoo.in</a>><br>
To: <a href="mailto:pw_forum@pwscf.org" target="_blank">pw_forum@pwscf.org</a><br>Date: Fri, 24 Jul 2009 17:06:10 +0530 (IST)<br>Subject: [Pw_forum] parallel execution of PWscf<br><table border="0" cellpadding="0" cellspacing="0">
<tbody><tr>
<td style="font-family: inherit; font-style: inherit; font-variant: inherit; font-weight: inherit; font-size: inherit; line-height: inherit; font-size-adjust: inherit; font-stretch: inherit;" valign="top"><div>Hi </div>
<div>I am using LG Quad core processor machine with PWscf running on linux. I want to know whether there is any command to run it parallely on the four processors or by default with simple installation (i.e., ./configure and make all) it uses all the four processors?</div>
<div> </div>
<div>Neel</div>
<div>University of Delhi</div></td></tr></tbody></table><br>
<hr size="1"> </blockquote></div><br>