Why is CAM5 so much slower than CAM4?

[sean_m_davis@noaa_gov]

Hi, I've noticed really large differences between the speed of CAM5 vs. CAM4. CAM5 is approximately a factor of 3-4 slower than CAM4 in testing I've done. I've tested with the compsets F_2000 and F_2000_CAM5.  I'm surprised that the differences are as large as they are, and was wondering if someone could confirm whether or not this behavior is seen on other systems. Also, are there namelist options in CAM5 that can be turned off to dramatically increase the speed (e.g., aerosols)? Thanks!

 

[santos]

A difference of 2-3x is not unusual; I personally haven't seen a difference as high as 4x before. I believe that the number of advected constituents in CAM5 contributes a substantial part of the difference, so MAM3 does play a role.

 

[andrew]

Hi Sean, Yes, the behavior is expected.  2x is CAM physics itself (new clouds: microphysics and macrophysics, and radiation). 2x is the Prognostic Modal Aerosol Code. About 2/3 of these changes are due to the physics, and 1/3 to the addition of advected species for clouds and aerosols.  You can turn off the modal aerosols. I am not sure the prescribed modal aerosols have been released yet. This yields about a factor of 2 speedup.You could use prescribed bulk aerosols (-chem none). Note that is not a fully supported configuration: the clouds will need some tuning. And it is not recommended if you are looking at cloud fields or cloud microphysics. Hope that helps.  Andrew

 

[sean_m_davis@noaa_gov]

Andrew, Sean, Thanks for your replies. That is very helpful. It sounds like using prescribed modal aerosols is the most desirable way to gain a factor of 2 for my purposes, since I don't have an interest in cloud tuning. This document leads me to believe that prescribed MAM has been implemented already -- http://www.cesm.ucar.edu/management/CSL/CSL12-14AccomplishmentsCESMFinal.pdf Is it simply a matter of changing the atm namelist variable prescribed_aero_model=MODAL in order to turn on prescribed MAM ?    

 

[andrew]

Hi Sean, I think so. I have not used it, and I think it is being updated, but prescribed MAM (modal) should work for your purposes if it has been released. Cheers, Andrew

 

[eaton]

If you're running the cam5 physics, then switching from prognostic to prescribed aerosols (via -chem none)  will default to prescribed modal aerosols.  build-namelist will set prescribed_aero_model='modal' by default in that case.We currently only supply pre-industrial and present day climatologies for the FV grid 1.9x2.5.  This is the only grid resolution that has been verified to produce a similar climate as the prognostic aerosols for PI and PD cases.  We have relied on PNNL for these datasets and the validation.  As Andrew mentioned they are still working on obtaining good simulations with prescribed modal aerosols, and have cautioned not to expect that the 2 deg datasets can simply be interpolated to other resolutions and give valid results.  They have also not yet provided datasets for use in 20th century transient simulations.