Re: [AMBER] NaN problem with GTX680

From: Jason Swails <>
Date: Tue, 3 Sep 2013 07:50:49 -0400

On Tue, Sep 3, 2013 at 7:43 AM, David A Case <>wrote:

> On Tue, Sep 03, 2013, JAVAD MAHMOUDZADEH wrote:
> > I am simulating a
> > polymeric system with nearly 80000 atoms including water molecules in a
> > truncated octahedron box. I am using GPU in the stage of production for
> > calculations on amber 12 with a GTX680-4G card. I passed the minimization
> > stages and heating using sander and a 50ps MD for density equilibration
> using
> > pmemd.cuda with success. In the stage of production, using pmemd.cuda,
> unfortunately
> > after around 29ns, the restart file became full of NaN and the density
> of the system
> > became 0.02 and I observed *********** in front of EPtot and BOND in
> .out file.
> Run a *much* shorter simulation, probably on a CPU. Watch the behavior of
> the
> density. It might help to visualize the resulting trajectory. Try to use
> mostly defaults in the input file (and/or post your mdin file). It seems
> likely that your problem occurs way before 29 ns, but without more
> information, people on the list cannot be of much help.

Also make sure you are using a completely up-to-date version of Amber 12.
 If you run the command

$AMBERHOME/update_amber --version

you should see the following output:

Version is reported as <version>.<patches applied>

AmberTools version 13.16
     Amber version 12.19

If the <patches applied> are less than 16 and 19, respectively, re-run the
configure command (saying 'yes' when asked about downloading updates) for
CUDA and recompile.

While the above is not debugging advice (follow Dave's advice there -- and
debugging is usually done on the CPU because it provides helpful error
messages), it is a useful practice to make sure you don't spend time
hunting down the root cause of a bug that has already been fixed.

Good luck,

Jason M. Swails
Rutgers University
Postdoctoral Researcher
AMBER mailing list
Received on Tue Sep 03 2013 - 05:00:04 PDT
Custom Search