On Fri, Nov 08, 2024, Rosa Teijeiro Juiz via AMBER wrote:
>
>I am writing to ask for help to optimize the time of my MD
>simulations. So far, I have only worked with short simulations because
>I was still learning how to use the software, but now I am trying to
>run a 500ns MD simulation of my protein and my problem is that the
>estimated time for the run (what appears on my .mdinfo file) in over
>8000 thousand hours, which honestly doesn't make sense to me.
>
>I tried submitting the MD run to a cluster (instead of just running on
>my local computer) to see if this would help with the optimization of
>the time but initially it was the same:
>
>AmberTools/23.6-foss-2023a/bin/sander -O -i 03_Prod.in -o 03_Prod.out
>-p solvated_cotb2.parm7 -c 02_Heat.ncrst -r 03_Prod.ncrst -x
>03_Prod.nc -inf 03_Prod.info &
To make things go faster, in ascending order:
1. use sander.MPI, with something like 16-32 cores, if you have them
available. Depending on the size of your system, this might give close to
an order of magnitude speedup.
2. use pmemd.MPI. This is likely to be roughly 2-3 times faster than
sander, at least for a lot of systems.
3. get access to a GPU, and use pmemd.cuda. This is likely to be 100 or
more time faster. (Don't bother with pmemd.cuda.MPI until/unless you get to
using techniques like replica exchange.)
A discussion of these options is in Chapter 2 of the Amber 2024 Reference
Manual.
...good luck...dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Nov 08 2024 - 16:30:02 PST
