[AMBER] possibly a bug in the CUDA version of PMEMD

From: Ye Mei via AMBER <amber.ambermd.org>
Date: Mon, 17 Mar 2025 11:56:43 +0800

Dear Amber Development Team,

I am writing to report an issue with an absolute binding free energy (ABFE)
calculation of a small molecule binding to β-cyclodextrin. During the
alchemical decoupling step in the binding pocket, the pmemd.MPI executable
completes the simulation successfully, but pmemd.cuda fails almost
immediately with the following error message:

of length = 42Failed an illegal memory access was encountered

I have attached the job file for reference. The structure was minimized
prior to this decoupling step, and I have tested it using both Amber20 and
Amber24 with the same result. This leads me to suspect a potential bug in
the CUDA version of pmemd.

Could you please advise on how best to resolve this issue, or let me know
if it is already known and being addressed?

Thank you for your assistance.

Sincerely,

Ye Mei

East China Normal University


_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber

Received on Sun Mar 16 2025 - 21:00:02 PDT
Custom Search