<html>
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
On 11/17/11 10:10 PM, Andrzej Rzepiela wrote:
<blockquote
cite="mid:97E0ABA1-7682-4C7E-8FF8-D98BFB1B0DAF@physik.uni-freiburg.de"
type="cite">
<div>Hey, </div>
<div><br>
</div>
<div>thank you for the hint. I just finished the tests. The
machine is Intel xeon R, 12 cores , 4 Teslas M2090 and 96 gb
of memory.</div>
<div>I used the most demanding PME dhfr benchmark system ( 7000w +
protein) and obtained the following results:</div>
<div><br>
</div>
1 cpu core run: 2.135 ns a day<br>
12 cpu cores run: 21.478 ns a day<br>
<br>
1 gpu unit run ( how many cpu cores are used, one ? ): 13.0
ns/day<br>
<br>
<div>4 parallel gpu runs on the node, performance: 12.884,
10.296, 8.290 and 6.953 ns/day, which is on average about 9.6
ns for one gpu.</div>
<div><br>
</div>
<div>From your experience, are this expected numbers ? From the
benchmarks on the website I had e feeling that the gpu runs will
be faster.</div>
</blockquote>
GPU is faster with implicit solvent and other constraints up to now.
They are optimizing everything for the next version of gromacs<br>
<a class="moz-txt-link-freetext" href="http://www.gromacs.org/Developer_Zone/Roadmap/GROMACS_4.6">http://www.gromacs.org/Developer_Zone/Roadmap/GROMACS_4.6</a><br>
<br>
More detailed info on the gromacs manual.<br>
<blockquote
cite="mid:97E0ABA1-7682-4C7E-8FF8-D98BFB1B0DAF@physik.uni-freiburg.de"
type="cite">
<div><br>
</div>
<br>
<fieldset class="mimeAttachmentHeader"></fieldset>
</blockquote>
<br>
<br>
<br>
<pre class="moz-signature" cols="72">--
Gianluca Santoni,
Institut de Biologie Structurale
41 rue Horowitz
Grenoble
_________________________________________________________
Please avoid sending me Word or PowerPoint attachments.
See <a class="moz-txt-link-freetext" href="http://www.gnu.org/philosophy/no-word-attachments.html">http://www.gnu.org/philosophy/no-word-attachments.html</a></pre>
</body>
</html>