On Thu, Apr 29, 2021, 石谷沁 wrote:
>
>Thanks for your follow-up. If the errors are with GB calculation, then
>that's quite serious because the most common post-processing tasks we will
>do is MMPB(GB)SA.
Do you use pmemd.cuda for such tasks? Is it really significantly faster
than pmemd or sander on a CPU? I ask because getting single frame energies
involves a lot of initialization tasks that reduce the speed advantage of
the GB calculation itself.
>
>David, what will be the potential way to solve this DPFP
>errors? Re-compiling with different ways? Install on a different
>workstation/cluster.....?
Both of the above are good ideas. It's certainly possible that you are
seeing a hardware problem, so trying a different GPU might help.
Changing compilers or OS might also help identify the problem. Right now,
we don't have any way to reproduce the problem at our end.
...regards....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Apr 29 2021 - 05:30:04 PDT
