On Wed, Jan 06, 2021, Matthew Guberman-Pfeffer wrote:
>
>My system starts
>out the NPT equilibration with dimensions (129.600 32.400 101.100). The
>calculation proceeded until timing out on the HPC. When I restarted
>the calculation, the system had dimensions (128.374 32.093 100.143). I
>understood from the other post that the ~30 Å dimension is problematical
You are correct: it's the short dimension that is the problem. Here are the
options as I see them:
1. Add another layer of unit cells in the y direction, to get that up to
64 Å.
2. Try a smaller cutoff, but don't go below 8 Å, which is the default.
You may have to use some trial and error, to find the largest value that the
GPU will accept.
3. Use the CPU version of pmemd.
4. Try another code base: wikipedia has a nice summary:
https://en.wikipedia.org/wiki/Molecular_modeling_on_GPUs
I've used openMM quite a bit for crystal simulations, and have posted
some simple scripts: https://github.com/dacase/OMM_Amoeba_Scripts
These emphasize the Amoeba force field (since Amber's GPU code doesn't
support Amoeba). But you could change the force field to many other choices
that openMM supports. One advantage is that output trajectory and restart
files are in Amber/Netcdf format, so all of the analysis tools you may be
used to using from Amber are available.
My personal preference would be option 1.
...good luck...dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Jan 06 2021 - 06:30:05 PST
