Hi,
I noticed that the runtime of my UM suites (GC3.1/UM10.7) increased this month by about 25% compared to running the same suites in November of last year (see output at bottom). I was wondering whether this could be related to the CPU frequency on ARCHER2 that was reduced from 2.25 to 2.0 GHz on 12 Dec 2022?
I want to increase the CPU frequency again to test whether this is indeed the problem. The ARCHER2 documentation (Energy use - ARCHER2 User Documentation) recommends to add
‘export SLURM_CPU_FREQ_REQ=2250000’
to the job submission script. Could you point me to where I could add/change this in my UM suite (u-cn504)?
Thank you very much for your help.
Best regards,
Seb
Below is the information of two jobs (via ‘sacct’), each running one year of the same suite (u-cn504) with the same number of nodes.
09/11/2022:
JobID Elapsed ReqCPUFreq AllocNodes ConsumedEnergy
2583538 03:59:42 Unknown 8 55.65M
2583538.bat+ 03:59:42 0 1 7.44M
2583538.ext+ 03:59:42 0 8 55.65M
2583538.0+0 03:58:55 Unknown 6 14.89M
2583538.0+1 03:58:55 Unknown 1 7.34M
2583538.0+2 03:58:55 Unknown 1 3.72M
13/01/2023
JobID Elapsed ReqCPUFreq AllocNodes ConsumedEnergy
2897996 04:56:37 Unknown 8 50.41M
2897996.bat+ 04:56:37 0 1 6.66M
2897996.ext+ 04:56:37 0 8 50.41M
2897996.0+0 04:55:07 2G 6 6.59M
2897996.0+1 04:55:06 2G 1 6.73M
2897996.0+2 04:55:06 2G 1 4.31M
The more recent jobs takes ~1 hour longer, but still consumes less energy due to the lower CPU frequency. But I want to reduce the runtime again.