[Openmm-team] gromacs @ gpu

[Peter Eastman]

Hi Maik,

We are almost ready (within days, hopefully) to release OpenMM Preview  
Release 2.  It includes GPU code for an O(N^2) algorithm, as well as a  
version of Gromacs modified to use OpenMM.  We have a paper currently  
in press describing our results with it.  We have observed speedups of  
well over 100x compared to AMBER.  We haven't compared it to Gromacs,  
since it uses implicit solvent which Gromacs doesn't yet support.

We are also working on an O(N) algorithm so we can handle larger  
systems with explicit solvent.  We don't have any numbers for that  
yet, but hopefully it will also have quite large speedups.  It will  
probably be a few months before that is ready to release, though.

Peter


On Jan 20, 2009, at 4:24 AM, Maik Nijhuis wrote:

> Dear Openmm team,
>
> Currently I am investigating how to get the maximum performance out of
> Gromacs. On your website I see that you are using GPUs within  
> Gromacs. I
> am therefore very interested in your results, especially regarding  
> these
> items:
>
> - The Folding at Home project uses an n^2 algorithm in its Cell version.
> Although it seems less efficient than the original n log n algorithm  
> in
> Gromacs, it is more accurate and it reduces the complexity of the
> implementation, which improves performance. Do you also use a modified
> version of Gromacs, with an n^2 algorithm?
>
> - Could you tell me some (preliminary) results, for example the
> performance gain of your GPU version relative to the CPU version of
> Gromacs?
>
> Regards,
>
> Maik Nijhuis
> -- 
> maik at cs.vu.nl || http://www.cs.vu.nl/~maik
> _______________________________________________
> Openmm-team mailing list
> Openmm-team at simtk.org
> https://simtk.org/mailman/listinfo/openmm-team