Gave an updated talk at #HEPiX on carbon lifecycle analysis for scientific computing exploring how long you should run compute nodes when optimizing for minimizing carbon emissions.
We use real numbers from four data centers, and a couple of hypothetical cases, to explore how the local electricity generation emissions impact how sensible it is to run old hardware until it breaks.
Also tagging my co-author @wim_v12e who has done most of the model design.