VMD-L Mailing List

From: John Stone (johns_at_ks.uiuc.edu)
Date: Tue Jul 13 2010 - 14:50:08 CDT

Hi,
  Regarding the effective speedup for an 8600M GT:
It depends greatly on which GPU accelerated features of VMD
are being used. With only 32 SPs, at 950Mhz, its going to be
around 15% of the speed of an 8800GTX, but all of the CUDA kernels
currently in VMD get speedups of 20 or so for noteworthy problem
sizes on 8800GTX type hardware. Cutting this by a factor of 10,
there might still be a usable speedup depending on which kernels
one is running. Also, even if it's not much faster than the
CPU, the GPU calculation could end up being more power efficient,
which is still an interesting scenario for laptop use.

Thus far I haven't had an opportunity to carefully investigate the
impact of CUDA/OpenCL GPU-acceleration on laptop battery life for
the algorithms in VMD, but the benefit would depend on the
GPU speedup vs. additional power consumption.

In terms of the impact on the graphics performance, if the GPU kernels
are launched from within VMD, there's no performance loss during
interactive rendering of things like molecular orbitals, even though
the GPU ends up being used for both calculation and for display, since
they are effectively time-sliced. The case where one would be concerned
for impact on graphics performance is when one would launch a long-running
batch type simulation analysis calculation. That would be a case where
interactive graphics performance would definitely take a hit, while the
batch mode number crunching is going on.

There will be progressively more CUDA acceleration in VMD, some of the
things I have sitting on the back burner include faster molecular surface
display and other visualization-related kernels, where the GPU should
be giving us speedups of 20 or more (from early versions of the code)
and will make various types of interactive visualizations much more
responsive. I hope to make progress on that code again in the near
future, I expect people will like that a lot.

I am indeed interested in collecting user feedback on the things they
want to see run faster in VMD, whether by GPUs or otherwise.
We'll likely be conducting a user survey in the next couple of months
to assess some of this, but email feedback is always welcome.

Cheers,
  John Stone
  vmd_at_ks.uiuc.edu

On Mon, Jul 12, 2010 at 05:50:26PM -0400, Axel Kohlmeyer wrote:
[...]
> > to install anything else?? Actually I have the same problem with a
> > MacPro computer with MacOSX and a cuda capable Graphic card....Please
> > help.
>
> i don't know about macos x, but first of all you should know
> that only a very limited subset of functionality in VMD is currently
> accelerated by CUDA and also there is not too much speedup (if at all)
> to be expected from cards like the aforementioned GeForce 8600M GT card.
> it will be at least one order of magnitude less than with a high-end
> GPU, and if this is the only GPU in your system, it will be mostly busy
> with producing graphics in the first place.
>
> what kind of application within VMD do you hope
> to accelerate with CUDA?
>
> that kind of feedback (from everybody) would be useful
> when deciding what part of VMD would be most attractive
> to add CUDA support to next.
>
> cheers,
> axel.
>
> >
> >
> > Raúl Araya Secchi
> > B.Sc Molecular Biotechnology.
> > Molecular Biotechnology Engineer.
> > Computational Biology Lab (DLab)
> > Center for Mathematical Modeling (CMM)
> > Facultad de Ciencias Físicas y Matemáticas.
> > Universidad de Chile.
>
> --
> Dr. Axel Kohlmeyer akohlmey_at_gmail.com
> http://sites.google.com/site/akohlmey/
>
> Institute for Computational Molecular Science
> Temple University, Philadelphia PA, USA. 

-- 
NIH Resource for Macromolecular Modeling and Bioinformatics
Beckman Institute for Advanced Science and Technology
University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
Email: johns_at_ks.uiuc.edu                 Phone: 217-244-3349
  WWW: http://www.ks.uiuc.edu/~johns/      Fax: 217-244-6078