[gmx-users] Confusing gmx tMPI Error

[William Barber]

Hello, I have a system which I'm trying to simulate. I can get it to run on
some systems with GPU or CPU just fine, and on one system which was working
fine with Gromacs a few days ago I can only run it on the CPU. If I run it
on the two GPUs in the system I get this error:

Wrote pdb files with previous and current coordinates
tMPI error: Receive buffer size too small for transmission (in valid comm)
Aborted (core dumped)

Clearly onl the tMPI line is truly relevant here. It's a very confusing
error to receive, and I've been unable to find anything on the web
regarding this error. The system functions fine on other systems on one or
more GPUs or on any systems CPU, so I can only assume it has something to
do with the GPUs on this system despite them working just fine a couple of
days ago. The only "errors" before that are LINCS warnings, which I expect
to see during this energy minimization process.

To rule out it being a damaged install of Gromacs I've also gone and
deleted and recompiled from scratch a clean copy of Gromacs 2016.4. Any
suggestions on how I can better diagnose or solve this problem would be
greatly appreciated! Thanks,

Will
University of Ontario Institute of Technology