Hi Antje,
Without wanting to confuse you I should probably just add the additional
caveat that the GPU code is still considered experimental. There are
differences in the precision model used so while we have tried to work hard
to make things directly comparable you should be aware of this when running
such calculations.
The CPU code is well tested though so AMBER 9 to AMBER 11 is perfectly
reasonable for direct comparisons.
All of this of course assumes that your system is stable and well behaved to
begin with. If you have highly strained atoms in your system then all bets
are of course off.
Good luck,
All the best
Ross
> -----Original Message-----
> From: Ross Walker [mailto:ross.rosswalker.co.uk]
> Sent: Friday, April 08, 2011 11:56 AM
> To: 'AMBER Mailing List'
> Subject: Re: [AMBER] Compare simulations done with AMBER9 to ones done
> with AMBER11 Cuda?
>
> Hi Antje,
>
> Just to add some more to Jason's possibly over cautious email. There
> should
> be no problem with you comparing results between AMBER 9 and AMBER 11
> (CPU
> or GPU) as long as you are running the same simulation parameters. As
> Jason
> said though individual trajectories will diverge. You can even prove
> this to
> yourself by running a simulation with AMBER 9, stopping it and getting
> the
> restart file and then running the 'restarted' simulation twice, once in
> serial and once in parallel and you will see things start to diverge.
> Neither is 'wrong' though, it is just rounding differences that cause
> things
> to take different paths through phase space. The same will happen with
> AMBER
> 11 however you should be perfectly fine to interchange amber 9 , amber
> 11
> CPU and amber 11 GPU within a single run. Just be aware that ultimately
> you
> need to run your simulations long enough that your results converge.
> This is
> irrespective of what version of the code you are using.
>
> All the best
> Ross
>
> > -----Original Message-----
> > From: Jason Swails [mailto:jason.swails.gmail.com]
> > Sent: Friday, April 08, 2011 11:38 AM
> > To: AMBER Mailing List
> > Subject: Re: [AMBER] Compare simulations done with AMBER9 to ones
> done
> > with AMBER11 Cuda?
> >
> > If you run for long enough (i.e. any amount of time worth writing a
> > paper
> > about ;)), then the results will diverge due to the chaotic nature of
> > many-body systems and the limits of machine precision. Thus, the
> only
> > things you can directly compare are ensemble properties averaged over
> > an
> > entire simulation.
> >
> > My suggestion would be to re-run some of the simulations you ran with
> > Amber9
> > with Amber11 CUDA and check that you get the same ensemble results.
> > This
> > should be a good indication that you can safely move throughout the
> > two.
> >
> > Hope this helps,
> > Jason
> >
> > On Fri, Apr 8, 2011 at 2:45 AM, Antje Wolf
> > <antje.wolf.scai.fraunhofer.de>wrote:
> >
> > > Dear list,
> > >
> > > I did some simulations with AMBER9. I would like to do an
> additional
> > > simulation with different conditions. As I am running out of time
> > with my
> > > study, I would like to use AMBER11 CUDA. Now I have some doubts if
> I
> > can
> > > compare the results later on. What are your experiences with
> > comparing
> > > results obtained with different AMBER versions?
> > >
> > > Your opinion is highly appreciated.
> > >
> > > Cheers,
> > > Antje
> > >
> > > --
> > > Antje Wolf
> > >
> > > Fraunhofer-Institute for Algorithms and Scientific Computing (SCAI)
> > > Department Bioinformatics
> > > Schloss Birlinghoven
> > > D-53754 Sankt Augustin
> > >
> > > Tel.: +49 2241 14 2552
> > > E-mail: antje.wolf.scai.fraunhofer.de
> > > Internet: http://www.scai.fraunhofer.de
> > >
> > > _______________________________________________
> > > AMBER mailing list
> > > AMBER.ambermd.org
> > > http://lists.ambermd.org/mailman/listinfo/amber
> > >
> >
> >
> >
> > --
> > Jason M. Swails
> > Quantum Theory Project,
> > University of Florida
> > Ph.D. Candidate
> > 352-392-4032
> > _______________________________________________
> > 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
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Apr 08 2011 - 13:30:02 PDT
