On Mon, Mar 21, 2022, Nathan Black wrote:
>
>To reduce the computational expense of a simulation for this large system,
>I attempted to set a nonbonded interaction cutoff of 10.0 Angstroms (cut =
>10.0), choose a Generalized Born implicit solvation model (igb = 5), and
>run using GPUs (implementation with pmemd.cuda). However, I received an
>error message when attempting this simulation- I did some further digging
>and found that the nonbonded interaction cutoff must be set to the system
>size when igb > 0 while using pmemd.cuda.
There is no cutoff in GB on GPU's in Amber: cut has to be set to some value
greater than 999..
>
>I will experiment with reducing the size of the system, but are there any
>known workarounds to this? I ask that question knowing that this design was
>probably intentional.
Have you actually benchmarked your system on GPUs? It is certainly true
that Amber's GPU code can become uncomfortably slow for very large systems.
It can actually be cheaper (in terms of time per MD step) to use explicit
solvent.
There is unfortunately no alternative in Amber's GPU code.
....hope this helps....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Mar 21 2022 - 10:30:03 PDT
