[MITgcm-support] Choice of workstation CPU for running MITgcm (Christoph Stappert)

Senja Walberg senja.w at gmail.com
Tue Apr 9 14:35:45 EDT 2019


Hey Christoph,

I'm sure you'll get better and more thorough answers from other people
here, but I thought I'd say that I did some tests on how the MITgcm scales
and compiled the attached data, I'm not sure if the image will show up or
not. Essentially there was a falloff in performance gain past 128-256
cores. On the cluster we run on you can request full nodes or let the
processes distribute across many nodes, and we found that it ran faster
distributed across many nodes as it had access to memory which was unused
by other processes.
[image: image.png]
[image: image.png]
Profiling the code showed that memory bandwidth is still a restriction.

As for the other questions, I have no idea, but hopefully this information
can help.

Regards,
Senja
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.mitgcm.org/pipermail/mitgcm-support/attachments/20190409/b5975891/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 133613 bytes
Desc: not available
URL: <http://mailman.mitgcm.org/pipermail/mitgcm-support/attachments/20190409/b5975891/attachment-0002.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 558340 bytes
Desc: not available
URL: <http://mailman.mitgcm.org/pipermail/mitgcm-support/attachments/20190409/b5975891/attachment-0003.png>


More information about the MITgcm-support mailing list