Chat with: Everyone

MarkR_Leeds #2
3:18 PM
Adrian: will disk space be common with work and home

Alan Simpson
3:19 PM
There are separate disks on the KNL system, though the standard ARCHER /home amd /work should be cross-mounted, I believe.

MarkR_Leeds #2
3:20 PM
Does that mean a separate login node

MarkR_Leeds #2
3:21 PM
no prob

MarkR_Leeds #2
3:24 PM
ta

MarkR_Leeds #2
3:35 PM
Does this mean one cannot guarnatee the ode when a node is assigned to your job?

MarkR_Leeds #2
3:35 PM
guarantee mode , sorry

MarkR_Leeds #2
3:39 PM
Ta

Stephan Walter #2
3:42 PM
Will be there also a video of this session again? And it would be great if there would be some comment to the MPI process pinning within SNC-4 Flat mode, so that all processes are spread equal over the four NUMA domains and use in the first place the MCDRAM. THX

Clair
3:43 PM
Yes, we hope to have the video up tomorrow.

Stephan Walter #2
3:44 PM
ok thx

Harvey
3:44 PM
We are aware of how to do the binding but that may (or may not) be covered in this talk 

Stephan Walter #2
3:52 PM
Ok, I will have a look tomorrow

MarkR_Leeds #2
4:06 PM
All good here thanks Adrian

Stephan Walter #2
4:09 PM
I have seen the Broadwell results before and have had some discussions with colleagues and we don't understand the very good CASTEP results. Any suggestions from your side? Feel free to contact me by mail.

MarkR_Leeds #2
4:10 PM
I think CASTEP has a well developed FFT solver (vague recollection)

Stephan Walter #2
4:12 PM
hm.. ok I have to think about that