[gmx-users] Cuda support

[Mark Abraham]

Jack Shultz wrote:
> Has anyone used these binaries? How are they? Any recommendations on using GPUs?
> 
> https://simtk.org/project/xml/downloads.xml?group_id=161#package_id600

Those binaries are just about useless. From their README...

-------------
Gromacs-OpenMM Preview Release 2

Gromacs-OpenMM is a modified version of Gromacs 4 which uses OpenMM to 
perform calculations instead of its own computational routines.  This is 
a preview release, and should be treated as prototype software.  It 
currently supports only a small fraction of the features of Gromacs 4.

<snip>

Supported Features

Only a fraction of the features of Gromacs 4 are supported in this 
release of Gromacs-OpenMM.  Here are some of the most significant 
limitations to the allowed options in the MDP file.

The only supported integrators are "md", "sd", and "bd".  The "bd" 
integrator is not supported on ATI GPUs.

The only supported settings for the "constraints" options are "none" or 
"hbonds".

Neither cutoffs nor periodic boundary conditions are supported.  The 
"coulombtype" and "vdwtype" settings are ignored, as are the various 
cutoff settings.

The only supported value for the "tcoupl" and "pcoupl" settings is "no". 
  Temperature coupling can be achieved by using the "sd" or "bd" 
integrator.  Pressure coupling is not a meaningful concept without 
periodic boundary conditions.

Rigid water molecules are not supported.

As a result of these limitations, explicit solvent cannot usefully be 
simulated with this version of Gromacs-OpenMM.  Implicit solvent can be 
added to a simulation by including the following three lines in the MDP 
file.

implicit_solvent = GBSA
gb_algorithm = OBC
gb_epsilon_solvent = 78.3