There seems to be a problem that comes up frequently here, which is that PGI builds will give the following error about drv_in being missing, even though it is present in the run directory:PGFIO-F-209/OPEN/unit=99/'OLD' specified for file which does not exist. Here are a bunch of posts that contain this issue:http://bb.cgd.ucar.edu/comment/1006759#comment-1006759http://bb.cgd.ucar.edu/parallel-run-problemhttp://bb.cgd.ucar.edu/error-executing-cam50http://bb.cgd.ucar.edu/errors-executing-cesm104-after-successful-build I had thought that this was also an MPICH-specific problem, but that last thread mentions OpenMPI. It appears that I can also reproduce this issue on Frankfurt using PGI and MPICH with the CESM scripts (but not OpenMPI or using the CAM Makefile). When I reproduce the issue, I can verify that the current working directory is correct before mpirun, and that it matches the value of the PWD environment variable during the run itself. So it is not at all clear to me why the PGI runtime thinks that the file does not exist.It's still not really clear to me what triggers this, but based on the reports here, it looks like this happens either when there is a descrepancy between the number of processors requested and the arguments to mpirun/mpiexec, or when the mpif90 compiler wrapper is not being used (using pgf90 or pgfortran instead). Looking through these threads, only two users mentioned fixing this problem, and both mentioned changes to the run script...