CPU layout in ACCESS-OM2-1deg

COSIMA
, , , , ,
1

Hi COSIMA people,
I am running some simulations branched from the release of ACCESS-OM2 1deg ryf.

In this configuration, the CPU layouts are chosen so that the total number of CPUs is 241. (216 + 24 + 1). This means there are 47 unused CPUs, and when submitting the job on payu, you get this message:

payu: warning: Job request includes 47 unused CPUs.
payu: warning: CPU request increased from 241 to 288

Is there a particular reason why it was set up to use 241 cores? Should we not adjust slightly to make it 240 cores to save resources?

Regards,
David

3

Hi @dkhutch,

I think the person best placed to explain the reason for this rather inefficient choice is @aekiss, but if I remember correctly, there are some constrains on the number of cores allocated to each component.

The software transformation team at ACCESS-NRI is planning on revisiting some of these configurations at some point with the goal of improving performance and efficiency, but the timeline is still unclear.

1 Like
4

This is a known issue that should be fixed, but we never got round to it as our focus was on high resolution. Happy to take contributions :slight_smile:

There are fewer idle cores with 28 cores/node on normalbw if that works for you.

1 Like
8

As there is a github issue to track any progress on this, and to avoid having multiple places to update/follow, Iā€™m closing this topic.