[gmx-users] Segfaults when using Replica Exchange

[Thomas Allen]

I'm having an issue where GROMACS is generating segfaults when I try  
to run replica exchange simulations on a system of ~300,000 atoms  
using 20 temperature points and around 10,000 cores. This happens when  
using either the 2016.3 version or the 5.1.2 version of the code, and  
I've tried varying the number of cores a bit and turning off dynamic  
load balancing so far, but neither of these things prevent it from  
eventually encountering this error. I have been able to run much  
smaller REMD simulations of the same model, and I can run regular MD  
on it successfully, but I would like to use replica exchange for the  
production calculations to improve my sampling.

I haven't had much luck debugging this through the sysadmins on the  
computing side so I was wondering if anyone was aware of what might be  
causing the trouble on the GROMACS end. I did see some old bug reports  
for version 4 regarding similar issues but my understanding was that  
these had been fixed.

Thanks,
Tom