Open akshaysridhar opened 5 days ago
There is a pending issue following build 699, in which the same configuration fails on CPU but passes on GPU devices (possibly due to recent bycolumn
changes in ClimaAtmos / Coupler)
We kept the separate GPU and CPU configs for this run only because we want to run it on GPU for 300 days, but running on CPU for 300 days will take a very long time (30 days already takes 7 hours). We kept the CPU config runs at 30 days to avoid wasting compute resources
There is a pending issue following build 699, in which the same configuration fails on CPU but passes on GPU devices (possibly due to recent
bycolumn
changes in ClimaAtmos / Coupler)
This configuration has been failing on CPU and passing on GPU for a few weeks (e.g. see this build from two weekends ago), so it predates at least the bycolumn changes that were made in the coupler
We kept the separate GPU and CPU configs for this run only because we want to run it on GPU for 300 days, but running on CPU for 300 days will take a very long time (30 days already takes 7 hours). We kept the CPU config runs at 30 days to avoid wasting compute resources
Ah noted, the SYPD values are 0.3 or 1.4 for CPU/ GPU respectively, I'll revert this change.
amip_target_topo_diagedmf_cpu/gpu
yaml files: CPU/GPU is now controlled via environment variables so we do not need 2 yaml files with identical config arguments.