ACCESS-OM2-025 + WOMBAT performance and scaling on Gadi

COSIMA
1

Hello,

Would someone have the ACCESS-OM2-025 + WOMBAT performance and scaling on Gadi to be included in a NCMAS proposal please?

Thanks,
laurie

2

Hi @LaurieM

Here are the ACCESS-OM2-025-BGC (0.25°, 50 levels) performance figures from last year:
13.1 kSU/yr
6.7 yrs/day
1824 cores

There’s a scaling analysis for 0.25° in Kiss et al 2020 but that was on Raijin and without BGC. As far as I know nobody’s done a scaling study on Gadi with BGC at 0.25°. @paulleopardi did you do any tests at 0.25°?

3

HI @aekiss

I did my ACCESS-OM2 testing at 0.1 degrees with JRA55v13.
See The ACCESS-OM2 Performance Report, especially pages 26 to 29.

4

Revisiting this same question, has anyone done any scaling analysis of ACCESS-OM2-025 on gadi in the last year? It would be great to have scaling published alongside ACCESS-NRI model releases!

5

Update in case anyone else is wanting ACCESS-OM2-025 scaling for NCMAS, it sounds like @minghangli is going to look into this now. Thanks @minghangli!

6

Sorry for unearthing this “old” thread, but I’m curious if these more recent scalings on Gadi were published somewhere in the end? (And as an aside, where could I find similar recent scalings for 1deg and 0.1deg versions?)

7

I think this was discussed in a TWG meeting today

@dougiesquire, @manodeep?

8

Hi @Benoit. Unfortunately we don’t have any updated scaling information at this time.

We are about to do new releases of all the configurations, including new WOMBATlite configurations at all three resolutions. I’ll make sure to include the cost of each configuration in its README so it is a little more transparent.

9

Hi @Benoit - While ACCESS-OM2 is not included (but might be in the future), this repo contains scaling results for various ACCESS-NRI models. We will be adding scaling data for new models as and when possible.

Are you interested in the scaling data for future NCMAS-type proposals or more out of curiosity?

10

For NCMAS potentially, for my curiosity certainly, but mostly for comparing wall time and SUs required to spin-up “slowly evolving” tracers like age and water-mass fractions.