[CIG-MC] CitcomS and Intel Xeon E5-2600 processors

Matthew Knepley knepley at mcs.anl.gov
Sun Sep 28 13:05:25 PDT 2014

On Sun, Sep 28, 2014 at 2:55 PM, Max Rudolph <maxwellr at gmail.com> wrote:

> Dear All,
> I'm getting ready to buy some computer equipment and I'm wondering
> whether anyone has experience running CitcomS on the recenty-released
> higher core-count Intel Xeon E5-2600 v3 series processors (or their v2
> predecessors), specifically the 12-core chips. The 12-core CPUs can
> run 24 threads (i.e. MPI processes) efficiently, so in principle it
> seems like dual 12-core Xeons would make for a very cost effective
> computer to do small 3D calculations, post processing, etc. A system
> with 2 of these processors and 128 GB RAM goes for around $7500 right
> now. My question though is whether the memory bandwidth is going to be
> too much of a limitation to make this a worthwhile alternative to
> buying more nodes with fewer cores per node. The per-core memory
> bandwidth (2x 8.0 GT/s QPI links per cpu) is somewhat higher for the
> 8-core cpus than for the 12-core cpus (2x 9.6 GT/s QPI), so if memory
> bandwidth is the limiting factor, it may be a wiser decision to go
> with more nodes at reduced core count. Any input would be much
> appreciated.

Memory bandwidth is definitely the limiting factor. However, since the
incremental cost
of cores is generally small, and you may want to run things in the future
are compute limited or close to it (chemistry), its hard to turn them down.


> Regards,
> Max Rudolph
> _______________________________________________
> CIG-MC mailing list
> CIG-MC at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-mc

What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.geodynamics.org/pipermail/cig-mc/attachments/20140928/e7dfbabf/attachment.html>

More information about the CIG-MC mailing list