Updating CESM3 to latest intel compiler ( intel/2025.2.1 )

[jedwards4b]

I think that we want to update to the latest intel compiler before we release cesm3.  It replaces the intel fortran compiler ifort with a newone, ifx, and of course answers change.   We don't have a ect test for the new model and so I think that we will need to rely on scientists' evaluations to determine if there are any problems with the new compiler.   What are your thoughts on what we should test, how long it should be run and who will evaluate them?

[mnlevy1981]

I'm sure there are other tests to add to this list, but I'll start with two recommendations:

1. Re-run whatever CESM3 PI control run is considered the latest baseline in cesm_dev with the updated compiler; the baselines are typically in the 100+ year range, but maybe we could get away with rerunning the last 50 years with the compiler swap? I.e. if the baseline is 175 years, branch off the end of year 125 for the ifx run.

2. Run a single day with nstep output from all the components, and compare output from the first couple time steps. I've done that with some of the ocean packages (vertical mixing and BGC tracers), and my understanding is that chaotic nature of the physics doesn't kick in immediately so differences should be very close to round-off initially... but I could be wrong on that assumption.

[billsacks]

Thanks for starting this discussion, @jedwards4b . Can you remind me what the barriers are to running ECT with the new model? I'm wondering if there's any hope of getting an ECT running that would at least cover a substantial portion of the CESM3 configuration even if it doesn't cover the whole model.

[jedwards4b]

Maybe @allibco will address this, but the parameters used to determine the weights need to be reconstructed based on the new components and grids. They will be completely different for cam and for mom and I would think that they will also be significantly different for ctsm too.

[allibco]

@jedwards4b @billsacks Sorry for the delay. We can easily configure ECT for the new release to test CAM and CLM. We just may want to adjust some of the parameters a bit, but now we (thanks to @teopb) have a clear outline of how to do this. If you point us to a tag, we can figure out the params for you (after the new year).

As far as testing MOM, that is something that Alper and I are working on. It'll be a top priority for me in Jan.

[billsacks]

That's fantastic - thank you, @allibco . CESM3 is still evolving a bit scientifically, so we're not ready with a scientifically-final tag, but for something close in order to adjust the parameters, you can use the recently-made cesm3_0_beta07. Thank you!

[billsacks]

(@jedwards4b - since you've been more involved with this than I have, please jump in if you have any corrections to what I said.)