It’s important to understand the capabilities of the computer you’re running climate models on, if for no other reason than to determine how long your submitted jobs will take to complete. Here are some baseline numbers generated by running CESM1.1.1 at 1 degree resolution, under compset F_2000_CAM5, on both Cori (generated Jan 2016) and Edison (generated Aug 2015):
In both cases, hyper-threading was used. Though the use of 4 threads has not been tested on Edison, it has been on Cori. In this particular case, using 4 threads instead of 2 yields a 45% speed-up for about a 37% increase in cost, so it’s unlikely this is a good choice to make when running long experiments.
To implement this hyper threading, it’s possible to set NTHRDS_ATM = X in the env_mach_pes.xml file, where X is some integer.