Re: [AMBER] Maximum system size

From: David A Case <>
Date: Sun, 26 Jul 2020 20:57:50 -0400

On Sun, Jul 26, 2020, Jack Shepherd wrote:
>Making a big system for simulation with pmemd.cuda recently I got the
>error c must be smaller than 10^4 angstroms and nfft3 must be between
>certain values.

The maximum value for nfft1,2,3 is hard-coded in mdin_ewald_dat.F90 (see the
gridhi variable), and the maximum box size is in boxhi (as 1000. Ang, not
10000 Ang., and least as I read the code.)

I know that people have doubled the gridhi value, and you might experiment
with making gridhi and boxhi larger. I'm not sure how these values will
carry over to pmemd.cuda, but you can experiment: run short simulations with
CPU first, then make sure you get equivalent results on the GPU.

Once you get a very big system volume, the number of atoms is likely to become
more of a limitation than is the size of the box or the number of PME grid
points. Don't expect this to be straightforward: a cubic box of water
with sides of 1000 Ang. would have 33 million molecules, or 100 million atoms.
This is well beyond the expected range of Amber simulations, and generally
requires both specialized hardware and software.


AMBER mailing list
Received on Sun Jul 26 2020 - 18:00:05 PDT
Custom Search