I had a similar problem, where my system size was just over 100k atoms and
my aMD jobs would crash but normal MD with 2fs time-step and 4fs time-step
(with HMR) would run without any problem.
I am running the jobs on Quadro P5000.
Could the size of the system killing my aMD jobs??
On Monday, August 19, 2019, David Cerutti <dscerutti.gmail.com> wrote:
> Your GPU is very old, and the out of memory problem means just that: there
> is not enough memory on the card. If your system has more than a few
> hundred thousand atoms, this is going to happen on those old cards. A
> newer GTX-1080Ti, Titan-V, or RTX card will handle any system that tleap
> can generate (up to a few million atoms) just fine. If you have only 680
> residues, the issue might be that your complex is very spread out or
> unfolded.
>
> If you're stuck with that old card, there are still things you can try. A
> single 680 residue globular protein will typically fit in a water box of
> 150k atoms. You might be able to switch to an octahedral box (rebuild in
> tleap) and save a little bit of volume that way. Otherwise, look at how
> much water you're trying to place around the whole thing and how far the
> complex is separated at the beginning of the run.
>
> Dave
>
>
> On Mon, Aug 19, 2019 at 8:55 AM Giulia Paiardi <g.paiardi.unibs.it> wrote:
>
> > It's a Tesla K20c.
> > Could there be problem with compatibility?
> >
> > Giulia
> >
> > Il giorno lun 19 ago 2019 alle ore 13:11 Jacopo Sgrignani <
> > sgrigna.gmail.com>
> > ha scritto:
> >
> > > Dear Giulia
> > > Which hardware (GPU model) are you using ?
> > >
> > >
> > > Jacopo
> > >
> > > Il lun 19 ago 2019, 13:01 Giulia Paiardi <g.paiardi.unibs.it> ha
> > scritto:
> > >
> > > > Hi amber users,
> > > > I try to set up a production for accellerated MD using amber 14.
> > > > I'm following the concepts of tutorial 22 of amber because my complex
> > is
> > > > bigger than that of the example (680 residue vs 58 of the example).
> > > > I obtained a stable equilibration process but when I start the
> > > production
> > > > with the following script
> > > >
> > > > AVP dt=2.0fs with SHAKE, NPT aMD boost pot and dih
> > > > &cntrl
> > > > imin=0, irest=1, ntx=5,
> > > > dt=0.002, ntc=2, ntf=2, tol=0.000001,iwrap=1,
> > > > ntb=2, cut=12.0, ntp=1,igb=0,ntwprt = 0,ioutfm = 1,
> > > > ntt=3, temp0=310.0, gamma_ln=1.0, ig=-1,
> > > > ntpr=1000, ntwx=1000, ntwr=1000, nstlim=2000000,
> > > > iamd=3, EthreshD=11481,
> > > > alphaD=544,EthreshP=-2705594,
> > > > alphaP=204136,
> > > > /
> > > > &ewald
> > > > dsum_tol=0.000001,
> > > > /
> > > >
> > > > the resulting error using pmemd.cuda is
> > > > cudaMalloc GpuBuffer::Allocate failed out of memory
> > > > Someone can help me resolving with problem?
> > > >
> > > > Thanks
> > > > Giulia
> > > >
> > > > --
> > > >
> > > >
> > > > Informativa sulla Privacy: http://www.unibs.it/node/8155
> > > > <http://www.unibs.it/node/8155>
> > > > _______________________________________________
> > > > 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
> > >
> >
> > --
> >
> >
> > Informativa sulla Privacy: http://www.unibs.it/node/8155
> > <http://www.unibs.it/node/8155>
> > _______________________________________________
> > 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
>
--
Best Regards
Elvis Martis
Mumbai.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Aug 19 2019 - 10:30:02 PDT
