computing costs for UKESM/HadGEM3 on ARCHER2

1

Dear Ros and Grenville,

I am a Lecturer at UEA and am currently in the process of writing an ERC grant application. For this purpose, I require a cost estimate for supercomputing on the ARCHER2 system using UK Met Office climate models. I was wondering if you could help me check my current estimates in case that I overlooked anything major?

I believe the costs per CU would be £0.39 for an ERC grant on ARCHER2 and the CMS website here implies that UKESM1 demands around 165.7 kAU per model year. Graphs on the same website appear to imply that HadGEM3-GC3.1-LL would require only marginally smaller computing costs of maybe 140 kAU/model year. Is this correct? I would have expected a greater discrepancy.

So, assuming £0.39/CU, the cost for UKESM would be £28.59 per model year and £24.26 for GC3?

Would be great to receive your input,

Peer

2

Peer,

We’d typically accommodate you under NERC resources at the partner rate (£0.2/CU) given that your work is NERC aligned (it is isn’t it?)

You might want see http://cms.ncas.ac.uk/wiki/Archer2#Performance for UKESM costs on ARCHER2. You might want to run some tests with your particular model configuration if the costs are critical.

Grenville

3

Dear Grenville,

Thank you for the info. So, just to confirm, is the 73.3 CU (135kAU) stated on the website the best current estimate for UKESM to use? Do I also understand correctly that there is no such clear estimate for the HadGEM3-GC3.1-LL model version?

Yes, the work would be NERC-aligned but purely ERC funded. Would it be reasonable to work with a £0.2/CU rate for the proposal in any case, or should I conservatively assume the higher rate in the proposal, just to ensure that the simulations can go ahead, i.e. is the £0.2/CU rate virtually guaranteed if the research focuses on atmospheric science / earth system science?

Thanks again,

Peer