[gmx-developers] Gromacs & CUDA

[Erik Lindahl]

On Feb 5, 2009, at 3:35 AM, <selarin at tmblog.ru> <selarin at tmblog.ru>  
wrote:

> Hello, everybody!
>
> Is there version of Gromacs that use CUDA? CUDA is NVidia's  
> architecture
> for parallel data processing using NVidia's GPUs. Does it planned to
> develop such version?

Yes, but it is not yet ready for normal simulations.

You can search for "OpenMM" to find a specific version of Gromacs that  
can use GPU libraries for all-vs-all atom interactions, but obviously  
that only works for small systems.

Neighborlists is an interesting challenge that we're working on, but  
currently you want to stick to a CPU there - they will even be faster  
if you use multiple cores (not to mention that PME works...)

Cheers,

Erik
  
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20090205/528ccea3/attachment.html>