?: incremental dev of CM2 and ESM1.5 to get to ESM2 - eg biennial ENSO in CM2
BE: what about NCI system resources? Is this a major constraint on ambition?
AH: system resources (SU) are not the main constraint - the main constraint is speed, ie walltime ie parallel scalability
BE: new capacity has just come online; ~3yr timescale for more
TZ: core speed, not number of cores, would help
?: in terms of department role - could influence NCRIS and NCI resources i its clear this is needed for CMIP7
DE: new cores have new architecture - many cores per CPU - so may be faster - would take experimentation, but not a factor of 2
HR: system will be under pressure once CMIP7 production runs are underway
BE: also what data resources
BGF: Finland has 5M people but the worldsâs 3rd fastest computer - can we lift the bar
RH: can we spin up component-wise to speed it up eg spin up ocean separately
ME: ~5x slower than NCAR model - but due to fewer vertical levels
?: NCAR is going to 80 levels. But what we need is fast aerosols
TZ: need to map out options in terms of model speed
?: timeline: must consider dev time for benchmarking systems - currently doesnât exist - we really need this because we werenât able to do performance tests of all the physics options available - if we could run multiple sensitivity expts in parallel we could make a lot more progress,
?: e.g to solve biennial enso - many ~50yr coupled runs to test parameter sensitivity
?: as well as IC ensembles, could do parameter ensembles. Could put Machine learning onto it
Oops, didnât mean to delete this post: attempt 2: I have also heard that the UM atmospheric model is a significant factor as to why ACCESS runs much slower cf. NCAR.