Different number of CPU lead to different result

[scs_wy@yahoo_cn]

Hi everyone
I run CCSM3 for a few days and I find something strange.
my cluster is HP SuperDom.
I do two same control run with different CPU(64 and 104) number but keep the same inverse ratio of each modle(atm 25% ocn 50% ice 6% lnd 13% cpl 6%).I find there is apparente difference on some variable。For example the SST has 0.5-2℃ difference on the two tests.
I guess different reselutions(atm and ocn) lead to the difference.
somebody can tell me why?
thanks

 

[rneale]

CCSM3 release code is not bit-for-bit due to the ocean (pop) code not being reproducible on varying numbers of processors. When you say there are SST differences of 1-2K, if that is over a few weeks or months, then that is ok as it could be divergence of an ENSO event or meandering of the gulf stream. If this is for many years though then that is a concern.