[MITgcm-devel] JFNK new test experiment

[Jean-Michel Campin]

Hi Martin,

The testreport run over night are not so great regarding the new
test using JFNK (offline_exf_seaice.dyn_jfnk). 

The good new is that it pass the 2+2=4 test, and go through
strict compiler debug/check option.

But many tests fails with only 2 or 3 digits of matching (for uIce 
standard deviation, but other stats are not better).

Seems that it's not so sensitive to compiler/platform, but
whether MPI is used or not (e.g., on new acesgrid cluster, the only 
non-MPI test pass with 16 digits, but all MPI fails at 2/3 digits).

Can talk later on today (to see what we can try).

Cheers,
Jean-Michel