A correlated problem associated with phase changes is that on a multi-processor machine in which processor affinities are used to keep the science application processes \'away from each other\', the processor affinities are lost. On a processor that supports hyperthreading, a slab model may then find itself scheduled on the same physical processor as another model, thereby approximately halving both their speeds. [This has just happened to a PC of mine.]

I don\'t propose that there can or should be a fix for this, but am just noting for others in the same situation that they may profitably reset the affinities after slab model phase changes.

[Edit: The affinity problem can be fixed by setting the affinity on the hadsm3_5.06_windows_intelx86.exe process as well as the hadsm3_um_5.06_windows_intelx86.exe process. The new worker process started at the phase change will inherit its affinity from the other process, which continues through the phase change. The graphics still won\'t work.]