I like Ross's idea better--stick to GPUs if you possibly can!
On Wed, Mar 30, 2016 at 4:14 PM, Ross Walker <ross.rosswalker.co.uk> wrote:
> Hi Voytek,
>
> You can also edit mdin_ctrl_dat.F90 in the src/pmemd/src directory and
> change
>
> integer, parameter :: gridhi = 512
>
> to
> integer, parameter :: gridhi = 1024
>
> Then recompile and I believe this should just work. It will up the memory
> requirements in places but it should be fine. This should carry over into
> the GPU code as well although I didn't look in detail but I don't see any
> reason why a larger FFT shouldn't work. It may not be optimal from a
> performance standpoint but I think it shouldn't cause any issues.
>
> All the best
> Ross
>
> > On Mar 30, 2016, at 13:08, Kasprzak, Wojciech (NIH/NCI) [C] <
> kasprzaw.mail.nih.gov> wrote:
> >
> > Dave,
> >
> > Thank you very much for the advice. I will try the higher order and
> larger grid spacing equal in all
> > directions.
> >
> > Voytek
> >
> > Wojciech (Voytek) Kasprzak [Contractor]
> > Analyst Programmer,
> > Basic Science Program, Leidos Biomedical Research, Inc.
> > Frederick National Laboratory for Cancer Research, Frederick, MD.
> > (301) 846 5537
> > http://binkley2.ncifcrf.gov/users/kasprzak
> > ________________________________________
> > From: David Cerutti [dscerutti.gmail.com]
> > Sent: Wednesday, March 30, 2016 3:46 PM
> > To: AMBER Mailing List
> > Subject: Re: [AMBER] Proper nfft* parameters for a very long solvent box?
> >
> > Hi Voytek,
> >
> > My recommendation would be to try nfft1 = 512, and set order = 6. Then,
> > adjust nfft2 and nfft3 as appropriate so that the grid spacings are
> roughly
> > equal in all directions, to get about 1.44A as you said. Go for values
> > that are multiples of 2 and 3; 5 is OK but it sounds like your box is big
> > enough that whatever a factor of 5 gives you the next stop on the
> > Two-by-Three Express isn't far. I do not know if this will work on the
> GPU
> > implementation--last I checked the default of order=4 was enforced and
> > other settings were unavailable.
> >
> > If you run into other problems, this sounds like a pretty big system, so
> > NAMD or OpenMM might be a good thing to look into. If you use desmond, I
> > believe that code will let you do hybrid interpolation orders, so you
> could
> > do 1.44A spacing in X with 1.0A spacings in Y and Z, then set (6, 4, 4)
> > interpolation order.
> >
> > Generally, whatever you were doing with the default 4th order
> interpolation
> > you can do with 6th order interpolation and up to 50% larger grid
> spacing,
> > for the same level of reciprocal space accuracy. Take a real space
> > electrostatics cutoff of 9A or greater and you'll have no issues from the
> > PME.
> >
> > Dave
> >
> >
> > On Wed, Mar 30, 2016 at 2:14 PM, Kasprzak, Wojciech (NIH/NCI) [C] <
> > kasprzaw.mail.nih.gov> wrote:
> >
> >> Dear Amber Users,
> >>
> >> I am trying to run an SMD simulation mimicking an optical tweezer
> pulling
> >> experiment
> >> on an RNA structure. The solvent box required to achieve full chain
> >> extension is
> >> approximately 720 A long. Letting the program (Amber 14, pmemd.MPI) set
> >> the PME charge
> >> grid parameters leads to an error: "nfft1 must be in the range of 6 to
> 512"
> >> I can set the nfft* params in the &ewald section, but I can't specify
> >> anything more than 512,
> >> which leaves the X/nfft1 = 1.44, i.e. above the 1.0A recommended in the
> >> manual.
> >>
> >> Are there any work-arounds for this limit? Should I set the nfft2 and
> >> nfft3 to yield the
> >> same 1.44 A spacing for symmetry, or can those params be set to values
> >> yielding 1.0 A?
> >> What are the potential consequences of keeping the asymmetric spacing?
> >> Is 1.44 just suboptimal or very bad?
> >> Any comments and suggestions would be much welcome.
> >>
> >> Thank you in advance,
> >> Voytek Kasprzak
> >>
> >> Wojciech (Voytek) Kasprzak [Contractor]
> >> Analyst Programmer,
> >> Basic Science Program, Leidos Biomedical Research, Inc.
> >> Frederick National Laboratory for Cancer Research, Frederick, MD.
> >> (301) 846 5537
> >> http://binkley2.ncifcrf.gov/users/kasprzak
> >> _______________________________________________
> >> 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
>
>
> _______________________________________________
> 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 Mar 30 2016 - 13:30:05 PDT
