Specifically, two bug updates were added, specifcally rssafehs4.4 and gdr1f406. Also using optimised versions of advection and radiation. Lots of other small changes to job deck. This job is stored in swsvalde and represents a BRIDGE standard HadCM3-MOSES1. Previously tcsyc was run on bigblue on July 2009 and used 32 processors (8NS /4EW). The latter should not matter but has rarely been tested. In order to get it to work, I had to include an additional update (%24MODS/general/ofilter_mpp.mod) which does change the science a little. The results should be near identical to tcmfa which itself was tested against xbped which was compared to benchmark results from the Hadley Centre themselves. This should also be the same as tcsyb. You should also look at tcvci/tcvcj/tcvck which is the same model seteup but with different distribution of processors. tcvciis identical to tcsyc but uses 4NS, 8EW and also includes Faster MPP energy adjustment code (which made no noticeable impact on speed), tcvcj is as tcvci (uses 4NS, 8EW) but without the energy adjustment option, tcvck is as tcvci but also includes a couples of additional updates (timerupd_new.mod and nomsrest) which have no impact on science.
|