Dear AMBER users,
I emailed the reflector about an issue last month regarding ABMD flooding
RNA simulations in AMBER18 that were crashing with a GPU memory error.
Since then, I have attempted numerous run-based fixes for this issue to no
avail, including:
- Changing the bias energy resolution
- Running exact replicas of previously successful runs
- Using more basic collective variables
- Deactivating periodic boundary conditions
- Manually re-centering the RNA in the solvent box
I recently found out that the issue started shortly after our cluster's GPU
drivers were updated, leading me to conclude that the crashes are the
result of a memory leak somewhere in the ABMD implementation that only
affected newer drivers(I should note that only the flooding mode is
affected; umbrella and analysis modes ran fine). I was just wondering if
this was a known issue, and if upgrading to AMBER22 would be sufficient to
solve it? Any help/experiences that others have had with this would be
appreciated. In the meantime, I plan to just run a larger volume of
simulations for shorter durations to avoid exceeding whatever memory buffer
is overflowing.
Best,
Owen O'Reilly
--
Owen O'Reilly | He/him | SJU(USci legacy) Class of 2024
"Be kind, for every person you meet is fighting a hard battle."
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Mar 22 2023 - 20:30:04 PDT
