GROMACS version: 2024
GROMACS modification: Yes/No
I am using GROMACS 2024 and have been running simulations successfully on the same version for about a year. Until last week, my runs completed without any issues. However, now I keep getting segmentation faults / ‘core dumped’ errors. I have also tried running different systems, but the problem persists.
We would need more information to understand what is happening. What has changed? Different input systems? Different conformations? Different input parameters? Operating system upgrade?
When do the segfaults happen? At the start of mdrun? Or during the simulation, if so at which step?
Which 2024 are you using? Are you using GPUs?
Thank you for quick response.
My GROMACS build reports only ‘GROMACS version: 2024’ without a minor release number (e.g., 2024.1 or 2024.2) ,
Yes I am using a GPU.
The segmentation fault occurs during the NVT and NPT equilibration steps. By different systems, I mean different protein–ligand complexes; however, the simulation parameters are the same ones I have successfully used in my previous simulations.
If the errors happen quick enough, could you try running without GPU (-nb cpu)? If we are lucky that gives a clearer error message. It might also be good to add -pforce 5000 to detect whether there are large forces present.
The error does not occur immediately. In some cases, the NVT and NPT equilibration steps complete successfully when I rerun them multiple times. However, during the production run, the simulation crashes after a few seconds. I tried running with -pforce 5000, and the following error was reported
“step 2546 atom 2618 x nan nan nan force nan
Program: gmx mdrun, version 2024
Source file: src/gromacs/mdlib/sim_util.cpp (line 297)
Fatal error:
At step 2546 detected non-finite forces on 2618 atoms”
There are no higher forces printed at steps just before for the same atom?
The most probable cause is a some atomic clashes around this atom.
I have now reinstalled GROMACS again (version 2025.4), but I am getting the same error. Even just to cross-verify, I reran a simulation that had previously completed successfully before this error appeared, to check whether there might be an issue with my simulation system. However, I encountered the same error again, even though the same system had successfully completed a 100 ns simulation earlier.
Did you see my last question? Did you get more large force prints before the last one with nan?
No.. “‘78 atom 2105 x nan nan nan force nan
step 78 atom 2106 x nan nan nan force nan
step 78 atom 2107 x nan nan nan force nan
step 78 atom 2108 x nan nan nan force nan
step 78 atom 2109 x nan nan nan force nan
step 78 atom 2110 x nan nan nan force nan
step 78 atom 2111 x nan nan nan force nan
step 78 atom 2112 x nan nan nan force nan
step 78 atom 2113 x nan nan nan force nan
step 78 atom 2114 x nan nan nan force nan
step 78 atom 2115 x nan nan nan force nan
step 78 atom 2116 x nan nan nan force nan
step 78 atom 2117 x nan nan nan force nan
step 78 atom 2118 x nan nan nan force nan
step 78 atom 2119 x nan nan nan force nan
Program: gmx mdrun, version 2025.4
Source file: src/gromacs/mdlib/sim_util.cpp (line 316)
Fatal error:
At step 78 detected non-finite forces on 2119 atoms”
i executed on cpu with -nb cpu (25 ps for NVT).. and it executed successfuly. no error of core dumped
So no large forces at step 77 or earlier?
So all crashes you observed were with a GPU?
No, there were no large forces at step 77 or at any earlier step. Yes, all the crashes I observed occurred when running with the GPU.
If you are running on the same machine with the same drivers, the same GROMACS version and the same simulation system, I am starting to suspect that something is wrong with the GPU.