It's more complex than that. CUDA PMEMD assumes an atom can only see one
periodic image within its cutoff within a given neighborlist box, which
means you can have 1 x 1 x 1 simulation with up to 26 periodic images, and
it should work. But if your cutoff extends past a neighborlist box
somehow, failure will occur.
Does this even work correctly on the CPU? Or just not crash ergo does
magical mysterious stuff that leads to ???
Scott
On Wed, Nov 9, 2011 at 9:10 AM, Ross Walker <ross.rosswalker.co.uk> wrote:
> Hi Giovanni,
>
> > I am running a simulation on a single C2050 GPU.
> >
> > The system is quite unconventional and composed by 9 single strand DNAs
> > attached with a short tiol onto a surface of gold
> >
> > The system is created in order to have an equal spacing between each
> > ssDNA
> > (also in terms of periodic images).
> >
> > To do so the water box is created with 0 buffer in X and Y directions,
> > and
> > with opportune spacing along Z.
>
> I know about this problem but have had it low priority to fix since it is
> actually quite useful in identifying people who are running too little
> solvent. The issue occurs when the solute touches both sides of the box at
> once. This occurs in poorly solvated systems when they rotate relative to
> the box or extend during a simulation. As soon as the contact occurs the
> kReduceForces error occurs. This is because of some assumptions made in the
> code that people would not run such unphysical simulations. That said there
> seem to be quite a few people running, as you put it, unconventional
> simulations.
>
> I will try and look into this some more and see if I can figure out a fix
> for it.
>
> > My input file is below. (# indicates a comment just to make it clear).
> > equil SAM
> > &cntrl
> > imin=0,
> > irest=0,
> > ntx=1,
> > nstlim=5000,
> > dt=0.002,
> > ntc=2,
> > ntf=2,
> > cut=8.0,
> > ntb=2, #the run is actually an NVT,
> > ntp=1, #this trick is used to monitor
> > taup=999999999999999999999999999, #pressure and density
> > ntpr=1000,
> > ntwx=1000,
> > ntt=3,
> > gamma_ln=2.0,
> > temp0=300,
> > ntr=1,
> > ig=-1,
> > iwrap=1,
> > ioutfm=1,
> > /
> > fix surface
> > 10.0
> > RES 1 5445
> > END
> > END
> >
> >
> > END
>
>
> > I need your help, since I don't understand if the problem is related to
> > the
> > system or to some problem with the GPU.
>
> It is a known issue. Can you try a couple of things for me please and let
> me
> know if it still fails.
>
> 1) Try running a true NVT simulation. That is set ntb=1 and remove the
> taup.
> I know what you say is a trick to get the density and pressure but it runs
> through a lot of additional code and will slow your simulation down quite a
> bit vs running true NVT. Note though I still expect this calculation to
> fail
> in the same way.
>
> 2) Try running the simulation as NPT with a reasonable taup of say 1.0 or
> so. This I believe should work since the box size should expand to prevent
> the contact at the edge of the box. It may just give you something crazy
> with your system though since you are at the limits of the underlying
> model.
> Make sure you us anisotropic pressure scaling (ntp=2). I'd be interested to
> know if this works while 1 above fails.
>
> All the best
> Ross
>
> /\
> \/
> |\oss Walker
>
> ---------------------------------------------------------
> | Assistant Research Professor |
> | San Diego Supercomputer Center |
> | Adjunct Assistant Professor |
> | Dept. of Chemistry and Biochemistry |
> | University of California San Diego |
> | NVIDIA Fellow |
> | http://www.rosswalker.co.uk | http://www.wmd-lab.org/ |
> | Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
> ---------------------------------------------------------
>
> Note: Electronic Mail is not secure, has no guarantee of delivery, may not
> be read every day, and should not be used for urgent or sensitive issues.
>
>
>
>
>
> _______________________________________________
> 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 Nov 09 2011 - 13:00:04 PST
