Re: [AMBER] pmemd.cuda incomplete output file

From: Andreas Tosstorff <>
Date: Thu, 22 Dec 2016 12:39:27 -0300

I finally was able to get at least an error message for my problem by
running the job directly from the command line:

ERROR: Calculation halted. Periodic box dimensions have changed too much
from their initial values.
  Your system density has likely changed by a large amount, probably from
  starting the simulation from a structure a long way from equilibrium.

  [Although this error can also occur if the simulation has blown up for
some reason]

  The GPU code does not automatically reorganize grid cells and thus you
  will need to restart the calculation from the previous restart file.
  This will generate new grid cells and allow the calculation to continue.
  It may be necessary to repeat this restarting multiple times if your
  is a long way from an equilibrated density.

  Alternatively you can run with the CPU code until the density has
  and then switch back to the GPU code.

So far I tried a 50ps NVT eq before the NPT eq. Do you think a longer NVT eq
could help here?

-----Original Message-----
From: David Case []
Sent: Sunday, December 18, 2016 10:45 AM
To: AMBER Mailing List <>
Subject: Re: [AMBER] pmemd.cuda incomplete output file

On Sun, Dec 18, 2016, Andreas Tosstorff wrote:
> The error is not fully reproducible, meaning that the equilibration
> stops at different times. Sometimes after 90 ps sometimes after 200 ps.
> I set ntpr=100 and printed out stderr with "pmemd.cuda .... > out 2>error"
> but no error messages were printed neither in the error file nor the

First, how big is your system? pmemd.cuda has a limit on the range of
forces that can be fit into a fixed-point representation, and larger systems
have larger fluctuations that can violate this. Having a fully equilibrated
system would reduce the likelihood of this.

Having said that, pmemd.cuda jobs are supposed to be reproducible. (Perhaps
not after a fixed-point overflow.) But a job that runs for 200ps once,
should give identical results for that long every time, and not sometimes
crash after
90 ps. [Here I am assuming that you are re-running the exact job.] Have you
ruled out hardware failure, e.g by seeing the same thing on a different GPU,
and/or by noting that all your other simulations seem to run fine on this


AMBER mailing list

AMBER mailing list
Received on Thu Dec 22 2016 - 08:00:02 PST
Custom Search