You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
ClimaAtmos has scaling runs for the "aquaplanet chap" setup introduced in CliMA/ClimaAtmos.jl#2724. These use 1, 2, and 4 GPUs each, and all run with MPI. We should setup analogous runs in ClimaCoupler using the same atmos setup/config files.
note: atmos prints SYPD at end of simulation. We'll need this too to plot SYPD vs # GPUs
This also requires setting up a new buildkite pipeline for coupler performance testing, analogous to ClimaAtmos' Target GPU Simulations pipeline. This pipeline will be triggered manually when we want to do performance testing, unlike the automated shortrun and longrun pipelines. I've set it up here: https://buildkite.com/clima/climacoupler-target-gpu-simulations
This pipeline (now at https://buildkite.com/clima/climacoupler-cpu-gpu-benchmarks) is being used for our benchmark runs. We decided that right now it's more useful to test the performance of different configurations, rather than to test the performance across architectures every week. We can manually run scaling tests as needed, or add a new pipeline for them if we decide that will be helpful
ClimaAtmos has scaling runs for the "aquaplanet chap" setup introduced in CliMA/ClimaAtmos.jl#2724. These use 1, 2, and 4 GPUs each, and all run with MPI. We should setup analogous runs in ClimaCoupler using the same atmos setup/config files.
note: atmos prints SYPD at end of simulation. We'll need this too to plot SYPD vs # GPUs
This also requires setting up a new buildkite pipeline for coupler performance testing, analogous to ClimaAtmos' Target GPU Simulations pipeline. This pipeline will be triggered manually when we want to do performance testing, unlike the automated shortrun and longrun pipelines. I've set it up here: https://buildkite.com/clima/climacoupler-target-gpu-simulations
part of #390
The text was updated successfully, but these errors were encountered: