George Carr had mentioned having run ccsm on phoenix at T85 (presumably gx1v3, full components), using 208 processors. I recall that atm used 128 processors, but how many processors did each of the other components use, and what was the throughput in simulated years per computing day (or computing hours per simulated year)?

FYI, I am running that case now on our IA64 cluster and it is taking about 5 hours per simulated year using 260 processors (cpl=4,ice=16,lnd=16,ocn=96,atm=128).

Also, FV 2x2.5_gx1v3 (compset B) on same machine with configuration using 232 processors (cpl=8,ice=16,lnd=16,ocn=96,atm=96) takes about 1.6 hours per simulated year.

Note that T85 has 2.5 times as many meshpoints as 2x2.5. I notice too that dtime for FV 2x2.5 defaults to 1800, and for T85 defaults to 600. Rad. freq. is 2 for the FV case, and 6 for T85, so the actual radiation frequencies (in seconds) are the same. Does the EUL dycore do the dynamics and tracer advection at the physics timestep (600)?

The answer is yes, CAM uses the same time step for dynamics, physics, and transport at any resolution with the exception of the radiative

transfer calculation where it is configured to operate once per hour.

Phil

Thank you for the reply on timestep size. I would appreciate if Goerge Carr could reply to the first question raised - namely the configuration used for this case on phoenix (number of processors for each component) and the timing. Thank you.