got different results when using different processors

[bing_pu@jsg_utexas_edu]

Hi all,

I ran E_2000 component (CAM4-CICE_CLM4 with slab ocean model) on the same machine. The 10-year average equilibrium temperature (2xCO2) is about 291.33 K (annual mean) when using 264 processors. But when I used 192 processors it was 291.46 K. Other differences between the two simulations are the values of CICE_MXBLCKS and PE layout. Not sure what happened and why the differences are relatively large (~0.13K). Any help are deeply appreciated!!

 

[dbailey]

If don't turn on the bit-for-bit flag in the coupler and there is regridding (which there is for the E compset), you will get different answers on different processor counts. This does raise a bigger issue here. The E compset and SOM specifically require that you generate a SOM forcing file. This compset is only intended to run out of the box for testing purposes. This is outlined in the data model documentation:

http://www.cesm.ucar.edu/models/cesm1.0/data8/data8_doc/book1.html

Dave

 

[bing_pu@jsg_utexas_edu]

Thank you very much, Dave!

I notice that the CICE user guide mentioned there is known bug--"CCSM coupler may have incorrect values in grid cells that change from an icefree state to having ice during the given time step, or vice versa, due to scaling by the ice area." Do you think this related to the problem here?

Also, I tried two tests with BFBFLAG on for one year, using 264 and 192 processors, respectively. Monthly output shows that global mean surface temperatures are the same from January-June but has small differences from July-December. What does this mean?