I wonder if anybody has a per-CU estimate of the CO2 cost of running calculations on ARCHER2? (One CU is one node hour.) I had a play with the green algorithms project and think the best estimate comes out as about 400g CO2e per CU and for a reasonably heavy computational proposal that comes out with a terrifying number and I wonder if I’m anywhere near right.
I don’t have specific data for ARCHER2, but how many CPU cores is there per node?
I know some clusters have calculated a carbon footprint of around 1-2 gCO2e per CPU core-hour and 35-40 gCO2e per GPU-hour. But these are obviously very location dependent, so I wouldn’t be surprised if it was different for ARCHER2.
It’s 128 cores per node (two 64 core AMD EPYC 7742 processors) and 256 GB ram. No GPUs on the standard nodes. I think I used EPYC 7763 when I put this through the web interface…