Re: Running NAMD CUDA

From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Fri Nov 04 2011 - 15:16:45 CDT

On Fri, Nov 4, 2011 at 4:06 PM, Steven Cox <scox_at_renci.org> wrote:

> What do I need to do differently to run the NAMD CUDA binaries?
>

first off, you should try the latest release version.
and if those don't work, the latest nightly build.

> I've followed these instructions:
> http://www.ks.uiuc.edu/Research/namd/2.7b2/ug/node70.html
>
> But consistently get:
> FATAL ERROR: Number of VDW types exceeds CUDA LJ_TABLE_SIZE
>

this means, that your input is hitting a limit.
GPUs have some (rather small) buffers for fast
access of data and NAMD tries to utilize them.
but their size limitations stipulate that not every
input can utilize them. of course, people first
implement the fast variant, and then later worry
about making it more general.

using a configuration that works with the CPU version.
>

data manament and size limitations on a CPU
are very different than on a GPU.

>
> log:
> NAMD_2.8b1_Linux-x86_64-CUDA/namd2 +idlepoll config.conf
> ------------- Processor 0 Exiting: Called CmiAbort ------------
> Reason: FATAL ERROR: Number of VDW types exceeds CUDA LJ_TABLE_SIZE
>
> Charm++ fatal error:
> FATAL ERROR: Number of VDW types exceeds CUDA LJ_TABLE_SIZE
>
> This is in the list archive:
> http://www.ks.uiuc.edu/Research/namd/mailing_list/namd-l/14498.html
>
> But appears unanswered.
>

perhaps the message is too obvious.
in any case, i don't know for certain if this issue has been
dealt with, but the easiest way to find out is to not use
an obsolete beta version, but try the official release and
if that has a problem the latest nightly build.

axel.

> Thanks for your help,
>
> Steve
>
>

-- 
Dr. Axel Kohlmeyer
akohlmey_at_gmail.com  http://goo.gl/1wk0
College of Science and Technology
Temple University, Philadelphia PA, USA.

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:57:53 CST