At the same time, I would emphasize that from a scaling point of view less cores per node is better than lots of cores because processors on the same node share the same network link. Also frequency of CPUs on 2-socket nodes goes higher than 4+-socket configurations.<div>
<br clear="all">--<br>Szilárd<br>
<br><br><div class="gmail_quote">On Wed, Jan 19, 2011 at 10:30 PM, Mark Abraham <span dir="ltr"><<a href="mailto:Mark.Abraham@anu.edu.au">Mark.Abraham@anu.edu.au</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div text="#000000" bgcolor="#ffffff"><div class="im">
On 20/01/2011 3:33 AM, Maryam Hamzehee wrote:
<blockquote type="cite">
<table border="0" cellpadding="0" cellspacing="0">
<tbody>
<tr>
<td style="font:inherit" valign="top">Dear Szilárd,
<div><br>
</div>
<div>Many thanks for your reply. I've got following reply
from my question from <span style="font-family:arial,helvetica,sans-serif;line-height:16px">Linux-PowerEdge mailing list. I
was wondering which one applies to GROMACS parallel
computation (I mean CPU bound, disk bound, etc).<span style="font-family:arial;line-height:normal"> <br>
</span></span></div>
</td>
</tr>
</tbody>
</table>
</blockquote>
<br></div>
In serial, GROMACS is very much CPU-bound, and a lot of work has
gone into making the most of the CPU. In parallel, that
CPU-optimization work is so effective that smallish packets of
information have to be transferred regularly without much
possibility of effectively overlapping communication and
computation, and so a low-latency communication network is essential
in order to continue making effective use of all the CPUs. Something
like Infiniband or NUMAlink is definitely required.<br><font color="#888888">
<br>
Mark</font><div><div></div><div class="h5"><br>
<br>
<blockquote type="cite">
<table border="0" cellpadding="0" cellspacing="0">
<tbody>
<tr>
<td style="font:inherit" valign="top">
<div><span style="font-family:arial,helvetica,sans-serif;line-height:16px">>There
shouldn't be any linux compatibility issues with any
PowerEdge <br style="line-height:1.2em;outline-style:none">
>system. At Duke we have a large compute cluster
using a variety of <br style="line-height:1.2em;outline-style:none">
>PowerEdge blades (including M710's) all running on
linux.<br style="line-height:1.2em;outline-style:none">
<br style="line-height:1.2em;outline-style:none">
>What interconnect are you using? And are your
jobs memory bound, cpu <br style="line-height:1.2em;outline-style:none">
>bound, disk bound, or network bound?<br style="line-height:1.2em;outline-style:none">
<br style="line-height:1.2em;outline-style:none">
>If your computation is more dependent on the
interlink and communication <br style="line-height:1.2em;outline-style:none">
>between the nodes, its more important to worry
about your interconnect.<br style="line-height:1.2em;outline-style:none">
<br style="line-height:1.2em;outline-style:none">
>If Inter-node communication is highly important,
you may also want to <br style="line-height:1.2em;outline-style:none">
>consider something like the M910. The M910 can be
configured with 4 <br style="line-height:1.2em;outline-style:none">
>8-core CPUs, thus giving you 32 NUMA-connected
cores. Or 64 logical <br style="line-height:1.2em;outline-style:none">
>processors if your job is one that can benefit
from HT. Note that when <br style="line-height:1.2em;outline-style:none">
>going with more cores-per chip, your max clockrate
tends to be lower. <br style="line-height:1.2em;outline-style:none">
>As such, its really important to know how your
jobs are bound so that <br style="line-height:1.2em;outline-style:none">
>you can order a cluster configuration that'll be
best for that job.</span></div>
<div><font face="arial,
helvetica, sans-serif"><span style="line-height:16px"><br>
</span></font></div>
<div><font face="arial,
helvetica, sans-serif"><span style="line-height:16px"><br>
</span></font></div>
<div><font face="arial,
helvetica, sans-serif"><span style="line-height:16px">Cheers, Maryam<br>
</span></font><br>
--- On <b>Tue, 18/1/11, Szilárd Páll <i><a href="mailto:szilard.pall@cbr.su.se" target="_blank"><szilard.pall@cbr.su.se></a></i></b>
wrote:<br>
<blockquote style="border-left:2px solid rgb(16, 16, 255);margin-left:5px;padding-left:5px"><br>
From: Szilárd Páll <a href="mailto:szilard.pall@cbr.su.se" target="_blank"><szilard.pall@cbr.su.se></a><br>
Subject: Re: [gmx-users] Dell PowerEdge M710 with
Intel Xeon 5667 processor<br>
To: "Discussion list for GROMACS users"
<a href="mailto:gmx-users@gromacs.org" target="_blank"><gmx-users@gromacs.org></a><br>
Received: Tuesday, 18 January, 2011, 10:31 PM<br>
<br>
<div>Hi,<br>
<br>
Although the question is a bit fuzzy, I might be
able to give you a<br>
useful answer.<br>
<br>
>From what I see on the whitepaper of the
Poweredge m710 baldes, among<br>
other (not so interesting :) OS-es, Dell provides
the options of Red<br>
Had or SUSE Linux as factory installed OS-es. If you
have any of<br>
these, you can rest assured that Gromacs will run
just fine -- on a<br>
single node.<br>
<br>
Parallel runs are little bit different story and
depends on the<br>
interconnect. If you have Infiniband, than you'll
have a very good<br>
scaling over multiple nodes. This is true especially
if it's the I/O<br>
cards are the Mellanox QDR-s.<br>
<br>
Cheers,<br>
--<br>
Szilárd<br>
<br>
<br>
On Tue, Jan 18, 2011 at 4:48 PM, Maryam Hamzehee<br>
<<a href="http://mc/compose?to=maryam_h_7860@yahoo.com" target="_blank">maryam_h_7860@yahoo.com</a>>
wrote:<br>
><br>
> Dear list,<br>
><br>
> I will appreciate it if I can get your expert
opinion on doing parallel computation (I will use
GROMACS and AMBER molecular mechanics packages and
some other programs like CYANA, ARIA and CNS to do
structure calculations based on NMR experimental
data) using a cluster based on Dell PowerEdge M710
with Intel Xeon 5667 processor architecture which<br>
> apparently each blade has two quad-core cpus. I
was wondering if I can get some information about
LINUX compatibility and parallel computation on this
system.<br>
> Cheers,<br>
> Maryam<br>
><br>
> --<br>
> gmx-users mailing list <a href="http://mc/compose?to=gmx-users@gromacs.org" target="_blank">gmx-users@gromacs.org</a><br>
> <a href="http://lists.gromacs.org/mailman/listinfo/gmx-users" target="_blank">http://lists.gromacs.org/mailman/listinfo/gmx-users</a><br>
> Please search the archive at <a href="http://www.gromacs.org/Support/Mailing_Lists/Search" target="_blank">http://www.gromacs.org/Support/Mailing_Lists/Search</a>
before posting!<br>
> Please don't post (un)subscribe requests to the
list. Use the<br>
> www interface or send it to <a href="http://mc/compose?to=gmx-users-request@gromacs.org" target="_blank">gmx-users-request@gromacs.org</a>.<br>
> Can't post? Read <a href="http://www.gromacs.org/Support/Mailing_Lists" target="_blank">http://www.gromacs.org/Support/Mailing_Lists</a><br>
--<br>
gmx-users mailing list <a href="http://mc/compose?to=gmx-users@gromacs.org" target="_blank">gmx-users@gromacs.org</a><br>
<a href="http://lists.gromacs.org/mailman/listinfo/gmx-users" target="_blank">http://lists.gromacs.org/mailman/listinfo/gmx-users</a><br>
Please search the archive at <a href="http://www.gromacs.org/Support/Mailing_Lists/Search" target="_blank">http://www.gromacs.org/Support/Mailing_Lists/Search</a>
before posting!<br>
Please don't post (un)subscribe requests to the
list. Use the<br>
www interface or send it to <a href="http://mc/compose?to=gmx-users-request@gromacs.org" target="_blank">gmx-users-request@gromacs.org</a>.<br>
Can't post? Read <a href="http://www.gromacs.org/Support/Mailing_Lists" target="_blank">http://www.gromacs.org/Support/Mailing_Lists</a><br>
</div>
</blockquote>
</div>
</td>
</tr>
</tbody>
</table>
<br>
</blockquote>
<br>
</div></div></div>
<br>--<br>
gmx-users mailing list <a href="mailto:gmx-users@gromacs.org">gmx-users@gromacs.org</a><br>
<a href="http://lists.gromacs.org/mailman/listinfo/gmx-users" target="_blank">http://lists.gromacs.org/mailman/listinfo/gmx-users</a><br>
Please search the archive at <a href="http://www.gromacs.org/Support/Mailing_Lists/Search" target="_blank">http://www.gromacs.org/Support/Mailing_Lists/Search</a> before posting!<br>
Please don't post (un)subscribe requests to the list. Use the<br>
www interface or send it to <a href="mailto:gmx-users-request@gromacs.org">gmx-users-request@gromacs.org</a>.<br>
Can't post? Read <a href="http://www.gromacs.org/Support/Mailing_Lists" target="_blank">http://www.gromacs.org/Support/Mailing_Lists</a><br></blockquote></div><br></div>