Tuesday 11th February
Attendees: @paulleopardi , @clairecarouge , @inh599 , @MartinDix , @tiloz , @anton , @manodeep , @Aidan , @spencerwong , @ben , @micael , @tammasloughran , @Jhan , @TommyGatti
Apologies: everyone else
Land updates
- Martin: build for checking uninitialised variables and out of bounds. Some problems found in CABLE that will change results. Still on going.
- Ian: Pb of energy conservation over land in the coupled model. Did a great job improving the water balance though. Question around using ILAMB for QA: can we use the NRI ILAMB release? Need to setup a reusable workflow for MOPPeR.
- Ian: open question about what the river inflow is in ocean. Need some help from someone who knows the ocean model. Martin: runoff output in ocean and river outflow output in atmosphere. Anton: runoff field in MOM5 called runoff. See Andrew Kiss notes, Anton to follow up.
Ocean updates
Nothing of note
Release updates
Tommy: prerelease current workflow not working as expected for all. Using branches for spec references isn’t working. Workaround for now is to use a commit hash as reference for spack.yaml. You now need to push to the spack.yaml with the new commit hash as well as pushing your code changes.- Aidan: it is a longer workflow with 2 pushes instead of one
- Jhan: Happy with the workaround. Makes it easier to determine which commits are included in a run.
Optimisation updates
- Manodeep got the configurations he needs. Trying to compile with OneAPI. Aidan: experience with MOM5, never a good idea to go to -O3, can be slower and incorrect. But that was a long time ago. Martin: need to use fpe-model= precise but never investigated in details. Manodeep: criteria is the same run needs to be deterministic. Micael: systematic testing this time and keep record of results for an informed decision. Aidan: load balance often as importance as any model optimisation. Manodeep: using the “sapphirerapids” queues might be a low hanging fruit to improve performance.
- Anton: Are updates to dependencies on the card (netcdf, mpi,…)? Manodeep: with OneAPI always trying the latest versions of dependencies first. Aidan: target specific dependencies but not all at once. Paul: dependencies on hh5 or access modules that could be replaced.
Actions
- Ian, Claire and Romain to discuss MOPPeR.
- Anton to follow up on river inflow in MOM5 with Ian.
Closing Remarks
- Meeting next week, Chris will be chairing.