I have a question about a notable discrepancy observed in two 10-year aquaplanet simulations.

The initialization files (ICMGG<expid>INUA, ICMGG<expid>INIT, ICMSH<expid>INIT) and fort.4 used were the same. The only difference between the simulations is that one was run on a cluster using the Cy43r3v1 source code, while the other used Cy43r3v2. However, according to the release notes of Cy43r3v2, both versions are supposed to be identical. They also had reliable acceptance tests. So, what could potentially account for the approximately 0.5 K and 1 m/s differences in zonal and time mean values?

Best regards,
Thomas