GROMACS version: 2022.5
GROMACS modification: No ( but I have installed it to use Cuda toolkit 12.1)
System: Linux Mint 21.2 OS, AMD 3970X CPU, Nvidia 2070 Super
After trying to run a 350ns simulation, I was having random error occasionally, where GROMACS states that “One or more water molecules can not be settled. Check for bad contacts and/or reduce the timestep if appropriate.” I am assuming that a water molecule is getting outside the box and violating the PBC conditions.
After changing the thread count, what is routed to the GPU, I thought it was an issue with AMD or something with my system, as I never had this issue on an Intel processor. So after reducing my thread count to my number of cores, and reducing the GPU tasks to PME, I was able to get a successful run. I tried running it again, the only difference being that I previously had -nt set to 32 and noticed a lot of processes switching to different cores so I set -pin to On instead and got the same PME issues. I am not sure why I am getting these seemingly random PBC errors. I tried adjusting my timesteps, but that really hasn’t solved my issue. Right now I am updating all my outputs every 50000 steps or 10 ps, pretty much the standard from what I have seen.
I have run some similar simulations on our campus cluster with these same or extremely similar files without issue, which leads me to think something is wrong with my local installation.
Any help or advice would be greatly appreciated