Dear Colleagues,
I experienced issues with QM/MM-TI calculations using the "sander.MPI + QUICK API" interface, even with no GPU involved, as I am trying to keep problems simple at this time.
For whoever is interested, please find the link below to reproduce the issue. The command to run is “mpirun -n 4 sander.MPI -ng 2 -groupfile groupfile”:
https://drive.google.com/file/d/1PsdqNEvXumDc58jCzRmP25InXUWl6-GF/view?usp=share_link
Symptom-wise, the job is stuck for at least 12 hours (and remains stuck).
Input-wise, the "groupfile" points to the "mm.mdin" and "qm.mdin" files, which control the QUICK API QM/MM-TI (basically, lambda=0 is mm, lambda=1 is qm). I have tried using PM3 and XTB to replace QUICK in the "qm.mdin" file, and they both worked fine without being stuck.
My gut feeling is that when QUICK API tries to register for resources, a race condition or deadlock occurs between the two MPI processes. But I could be wrong. Thank you for any insights to fix this issue.
The same issue also exists for sander.quick.cuda.MPI, but that can be resolved later. Please see the link below for a previous ticket at the QUICK GitHub.
https://github.com/merzlab/QUICK/issues/392
Best regards,
Zhen.
_____________________
Zhen Li<
http://lizhen62017.wixsite.com/home>, Ph.D.,
The Merz Research Group<
http://merzgroup.org>,
Michigan State University,
Cleveland Clinic.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Sep 24 2025 - 07:30:03 PDT
