Date: Thu, 23 Jul 2026 19:54:15 +0200
Dear Adrian,
Thank you very much for your quick reply.
Please find the answers below.
1. What programs did you use when running the heating stage?
The heating stage was run with pmemd.cuda.
The protocol was:
restrained minimization (pmemd.cuda)
intermediate minimization (pmemd.cuda)
unrestrained minimization (pmemd.cuda)
heating (pmemd.cuda)
The heating stage completed successfully and wrote the restart file
without any warnings or CUDA errors.
2. What parts were NPT and which NVT?
The protocol was:
Minimizations: no thermostat/barostat
Heating: NVT
First equilibration: NVT
Remaining equilibration stages: NPT
The crash occurs at the beginning of the first NVT equilibration,
immediately after the heating stage.
3. Try what you call "equilibration stage" with nptr=1 and see if it
ever takes one step or not.
I will test this and report back.
4. Try your equilibration stage with the CPU code for some steps, look
at it carefully, and see if that works, then move to the CUDA code.
I will also perform this test and compare the behavior with the CUDA version.
Did you patch the Amber code for the H200 cards?
No.
This is the standard Amber24 installation available on our HPC cluster.
The same Amber24 installation has already been used successfully for
another unrelated protein–ligand simulation on the same H200 GPU.
Thank you again for your help.
Also, find attached a snapshot of the files that are produced in my
molecular dynamic file, in the cluster
Best regards,
Adrian Roitberg via AMBER <amber.ambermd.org> escribió:
> Hi
>
> Two questions:
>
> 1. What programs did you use when running the heating stage ?
>
> 2. What parts were NPT and which NVT ?
>
> 3. Try what you call "equilibration stage" with nptr=1 and see if it
> ever takes one step or not.
>
> 4. Try your equilibration stage with the cpu code for some steps,
> look at it carefully, and see if that work, then move to the cuda
> code.
>
> Did you patch the amber code for the H200 cards ?
>
> Adrian
>
> On 7/23/26 7:35 PM, Eduardo Cueto Díaz via AMBER wrote:
>> [External Email]
>>
>> Following my previous email, those are more technical details I want
>> to share
>>
>> SYSTEM
>>
>> /Amber24 (pmemd.cuda)
>> NVIDIA H200 GPU
>> Driver 565.57.01
>> CUDA runtime 12.7
>> Linux HPC cluster/
>>
>> The simulated system is a protein–ligand complex (~57,500 atoms)
>> containing a bonded Zn/Mn binuclear metal center parameterized with
>> MCPB.py.
>>
>> The workflow is:
>>
>> /restrained minimization
>> unrestrained minimization
>> 100 ps heating/
>>
>> The heating stage completes successfully and produces a valid restart file.
>>
>> However, when starting the subsequent equilibration (ntx=5, irest=1),
>> pmemd.cuda aborts immediately with:
>>
>> /ERROR: AN ILLEGAL MEMORY ACCESS WAS ENCOUNTERED LAUNCHING KERNEL
>> KNLSKINTEST/
>>
>> The output reaches section 4 ("RESULTS") but no MD steps are performed.
>>
>> Additional observations:
>>
>> /TLEAP COMPLETES WITHOUT ERRORS.
>> THE TOPOLOGY APPEARS INTERNALLY CONSISTENT.
>> THE RESTART FILE IS VALID.
>> HEATING FINISHES NORMALLY WITH FINAL PERFORMANCE STATISTICS.
>> NO NANS, SHAKE FAILURES, OR VLIMIT EXCEEDED MESSAGES ARE OBSERVED.
>> THE SAME AMBER24 INSTALLATION HAS SUCCESSFULLY RUN ANOTHER UNRELATED
>> PROTEIN–LIGAND SIMULATION ON THE SAME H200 GPU./
>>
>> I searched the Amber mailing list archives but could not find reports
>> mentioning kNLSkinTest.
>>
>> MY QUESTION IS WHETHER THIS ERROR IS A KNOWN ISSUE IN
>> AMBER24/PMEMD.CUDA, PARTICULARLY FOR SYSTEMS CONTAINING
>> MCPB.PY-GENERATED BONDED METAL CENTERS, OR WHETHER THERE ARE SPECIFIC
>> DIAGNOSTICS THAT YOU WOULD RECOMMEND.
>>
>> Since the ligand is currently unpublished, I cannot distribute the
>> topology publicly. However, if necessary, I would be happy to provide
>> a minimal reproducer privately to one of the developers.
>>
>> Thank you very much for your time and assistance.
>>
>> Best regards,
>>
>> Eduardo Cueto Díaz
>> --
>> Dr. Eduardo J. Cueto Díaz
>> Instituto de Química Médica (IQM-CSIC)
>> Consejo Superior de Investigaciones Científicas (CSIC)
>> C/ Juan de la Cierva 3
>> 28006 Madrid, Spain
>>
>> Email: edujosecueto.iqm.csic.es
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>
> --
> Dr. Adrian E. Roitberg
> Frank E. Harris Professor
> Department of Chemistry
> University of Florida
> roitberg.ufl.edu
> 352-392-6972
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
-- Dr. Eduardo J. Cueto Díaz Instituto de Química Médica (IQM-CSIC) Consejo Superior de Investigaciones Científicas (CSIC) C/ Juan de la Cierva 3 28006 Madrid, Spain Email: edujosecueto.iqm.csic.es
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
(image/png attachment: Captura_de_pantalla_2026-07-23_195254.png)
