hassanbeydoun
hassanbeydoun
We decided CCN should be a mixing ratio in the SPA files. As for the preemptive multiplying by cloud fraction, that's done because nc is cell average while CCN is...
I was seeing slowdowns like this one back in April  But it got substantially worse recently. This is from a run on May 12: 
For reference, this is typical perf on pm-gpu:
> @hassanbeydoun have you had a chance to run those tests with ne30? I tested it in DP mode. It's BFB there.
@PeterCaldwell that's a good point. In the current code, qc2qi_hetero_freeze_tend is multiplied by il_cldm while nc2ni_immers_freeze_tend is multiplied by cld_frac_l so that's inconsistent. They should be multiplied by the same...
@crterai, yes that accretion prefactor is very low and we have seen instabilities occur at that lower end.
> Why are we running modal_aero anything anyways? Doesn’t spa bypass that stuff? Unfortunately SPA doesn't bypass all MAM calculations in v0. At some point we decided it wasn't worth...