Meeting minutes (24/06/2025)
@clairecarouge
@inh599
@Jhan
@Aidan
@anton
@dougiesquire
@ezhilsabareesh8
@manodeep
@micael
@paulleopardi
@RachelLaw
@spencerwong
@cbull
Chair: @clairecarouge
Minutes: @cbull
MOM5 and cmake
- Updating to MOM5 build with CMake (some context here). There is also updating the grid.
- Rachel: happy for changes to go in given we don’t expect any radical answer changes.
- Aidan (speaking for Harshula): does this affect generic tracers? Dougie: Currently, we are building from the development branch that Harshula has been looking into.
- Ian: are there any downstream impacts on the model? Land sea-mask changes? Dougie: Land sea-mask has not changed. Re-gridding weights have been changed a little.
- Claire: also supports bundling them together.
- Anton: was hoping to do the build and CICE5 changes at the same time. Are there other build changes planned for UM / Cable? Ian/Rachel: Yes, they are coming but they are a while away.
- Rachel: is it worth doing a 4xCO2 run in CICE5? Anton: is there a configuration already running with CICE4? Rachel noted that there is a config on GitHub but it doesn’t appear to be up to date with the spin-up configuration so Tilo had taken the dev-preindustrial+concentrations config and made the one line change required (changing CO2MMR value in line 652 in atmosphere/namelists). There is also some documentation in this forum post
- Ian: slight backtrack. Would be good to separate technical and science changes.
- @paulleopardi I missed your comment please add.
- @dougiesquire suggested some changes are ready now but others such as the grid change (relevant to the med’ salinity issue) are blocked, would be at least a week for that one
Updates on oneAPI compiler switch
See details in a table at the top comment in this PR
With the new compiler the atmosphere is now “faster” than the ocean (was the opposite using the old compiler), this has led to a different load balancing with the new compiler. (Micael clarified that MOM5 is actually just slower so the atmosphere is unchanged.) Manodeep is going to look into whether giving the ocean more cores will improve performance, this will break bitwise repro’ so he’s been delaying this test. Manodeep wanted to know if it was possible that we use the old compiler for spin up and then when we go to production with the new compiler. Yes, that is a possible solution. @RachelLaw (@inh599 agreed) suggested having ~100 years of the new one before going to production. @micael wished to emphasise the the old compiler will stop getting improvements but in theory the new compiler should be capable of the same performance and so will hopefully “get it” in the future (@cbull and @micael note that this change down the line could be answer changing.)
Decisions Made
- Proceed with infrastructure changes as they become ready, rather than waiting to bundle them. Separate technical and science updates where possible.
- Keep the spin-up configuration updated and documented on the forum.
- Avoid bundling too many changes at once to simplify debugging.
- Maintain flexibility in compiler use during spin-up, but aim to finalize before production.
- Accept that some answer changes may be inevitable with future compiler/machine updates.
Action Items
- Dougie to proceed with MOM5 and CMake updates when ready.
- Keep the spin-up configuration table updated on the forum.
- Rachel (?) to locate and share the forum post documenting the 4xCO₂ runs. - this is already linked in the comment above. I’ve added some extra detail around the config status.
- Manodeep to continue performance testing with oneAPI and explore optimal configurations.