VMD-L Mailing List

From: John Stone (johns_at_ks.uiuc.edu)
Date: Mon Apr 17 2023 - 11:13:20 CDT

Hi,
  Yes, as I described before, this isn't a question of recompiling or
software, this is purely a GPU memory capacity issue.
You can force VMD to run QuickSurf on the CPU by disabling VMD's use of
CUDA by setting the environment variable VMDNOCUDA to 1 prior to launching
VMD, e.g. according to your linux shell:
/bin/tcsh:
  setenv VMDNOCUDA 1
/bin/bash:
  export VMDNOCUDA=1

Ultimately, GPUs that only have 2GB of memory are going to be limited
in terms of how usable they are for QuickSurf or GPU-accelerated ray tracing.
Both of these are memory intensive, and the SLURM parameter you're passing
that is only asking for 1028MB should be raised to encompass 95% of the 2GB
of GPU ram on your P600 GPUs if you want to do anything with them.
Otherwise you're probably better off just telling VMD not to use them
since they are memory-starved for the purposes VMD intends...

Best,
  John Stone
  vmd_at_ks.uiuc.edu

On Mon, Apr 10, 2023 at 06:08:42PM -0700, Chris Taylor wrote:
> Thanks John, it's some Quadro P600s with just 2Gb RAM each. I tried rebuilding with NVIDIA-OptiX-SDK-6.5.0-linux64 and, just to see, adding "--mem-per-gpu=1028" to my SLURM sbatch but I get the same thing with and without the restriction. Thanks
> Chris
>
> > On 04/09/2023 10:58 PM PDT John Stone <johns_at_ks.uiuc.edu> wrote:
> >
> >
> > Hi,
> > How much RAM do the GPUs you're running on actually have?
> > While I wouldn't expect a 1.5M-atom structure to be a big
> > challenge for QuickSurf, it does need some decent GPU memory,
> > and if you've restricted it to anything less than several gigabytes,
> > I would perhaps start out with no restriction at all and work back
> > the other direction.
> >
> > If your next goal is to do some GPU ray tracing, that'll need even
> > more memory, so I would again recommend letting VMD use all of the
> > available GPU memory...
> >
> > If this doesn't work out, tell me more about the hardware and I can
> > give you some things to try out.
> >
> > Best,
> > John Stone
> > vmd_at_ks.uiuc.edu
> > 

-- 
Research Affiliate, NIH Center for Macromolecular Modeling and Bioinformatics
Beckman Institute for Advanced Science and Technology
University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
http://www.ks.uiuc.edu/~johns/           
http://www.ks.uiuc.edu/Research/vmd/