From: Peter Freddolino (petefred_at_ks.uiuc.edu)
Date: Tue Jun 17 2008 - 14:43:55 CDT
I know Jim is not available for a few days right now, but should have
some input on the current limitations of the available namd/cuda code
(which is still under development) when he's back and available -- I
can't speak to the other limitations of the posted code (which is still
being developed). If you send me the config files that give you trouble
I can pass them along.
Michael Heymann wrote:
> Hi all,
> I am a bioinformatician interested in the CUDA technology, both on the
> algorithm side and on the performance side.
> I'am not a MD specialist, I'm actually totally new to NAMD, but I'm
> working in a lab where some people use NAMD. Therefore, I'm
> investigating the potential benefits that the "CUDA version" of NAMD
> provides. I'm aware that it's still an experimental version, and that
> it's not suited for in-production use.
> The exemple provided with NAMD/CUDA works well but when I tried to
> launch NAMD on our own files, I had some troubles, among others:
> - too many atoms per patches (seems a hardware limitation of CUDA ?)
> - simulation cell becoming too small
> - constraint failure in RATTLE algorithm
> Note that these dynamics work well on a non-CUDA version of NAMD.
> That's why I'm wondering if there are known bugs in NAMD/CUDA that can
> explain these errors. Moreover, is there still a development effort in
> integrating NAMD with CUDA and is it possible to obtain a more
> up-to-date version than the one available since july 2007 ? Maybe via
> cvs ?
> Thanks in advance if somebody can give me some advices.
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:49:35 CST