Can you send me the system in question. We're aware of an issue with SPFP
and minimizations that start in unrealistically high energy cases that
would never be seen in anything short of a runaway MD run. The temporary
workaround is to do minimizations in SPDP while we gather enough of these
cases to figure out how to handle them transparently (the best way would be
to do the first few steps in SPDP while those clashes exist but NVIDIA
makes that awkward).
Scott
On Wed, Sep 5, 2012 at 9:20 AM, Tim Travers <tstravers.gmail.com> wrote:
> Hello,
>
> I recently updated an Amber 12 install with the latest bugfixes (#23 for
> AmberTools, #9 for Amber). I'm now having problems during minimization for
> a protein system in an octahedral TIP3P water box, which is neutralized
> with Na+ atoms and has a total of ~139000 atoms. The minimization runs all
> 1000 steps specified, but the total energy doesn't decrease from the
> initial 6.1E+06. In particular, the VdW energy always stays at a very high
> value.
>
> I noticed that the RESTRAINT energy is always at 0.0, even though I set a
> restraint mask for heavy atoms. This is the mask I used: ' !(:WAT) &
> !(:Na+) & !(.H=) '. I know this mask works because I used it in
> successfully minimizing another smaller system (~56000 atoms) using the
> same Amber 12 setup and control scripts.
>
> Since the setup works for the smaller system but not the larger one, I was
> wondering if there has been a reported problem in system size when doing
> minimization over the GPU? I'm using CUDA 4.2, and can replicate this
> problem on a GTX580 and GTX680.
>
> Thanks,
> Tim
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Sep 05 2012 - 11:00:03 PDT
