AMBER Archive (2008)

Subject: Re: AMBER: AMBER minimization problems

From: Robert Duke (rduke_at_email.unc.edu)
Date: Sat Dec 20 2008 - 18:10:03 CST


Memory is probably still a problem if you ran on 2 cpu's, even if it is on a supercomputer node. The big scale-down in memory is associated with dividing up the atom workload, and especially the nonbonded pairlist and the charge grids / fft's in the reciprocal space calculations. So try running on say between 16 and 32 cpu's for a minimization with sander or pmemd - should make a huge difference.
Regards - Bob Duke
  ----- Original Message -----
  From: Lili Peng
  To: amber_at_scripps.edu
  Sent: Saturday, December 20, 2008 6:53 PM
  Subject: Re: AMBER: AMBER minimization problems

  Hi Ross,

  Apologies for the late reply. Thanks for giving this a shot on your machine..

  I think I'll pass on the 10-year wait :). I was actually running this using 2 processors on a supercomputer cluster, rather than on a desktop, so I don't think computer memory makes a difference(?). Your suggestions about setting ntpr=1 to and running it using pmemd is a good idea. Can I try using pmemd while increasing the number of processes when I run it on a cluster?

  In the meantime, I have resorted to running it in implicit solvent and the simulations seem to be proceeding just fine. As another alternative I'm thinking about coarse-graining the PEG polymer while retaining the explicit solvent molecules (coarse-grained as well, of course). What do you think?

  Thanks for the help,
  Lili

  2008/12/18 Ross Walker <ross_at_rosswalker.co.uk>

    Hi Lili,

    So I tried this out on my machine and have a number of comments. Firstly this is a HUGE system, 673,130 atoms. Running on 2 processors of my desktop this requires 2.6GB of memory, so unless the machine you are running this on has around 4GB of memory the code will be swapping like crazy which means it likely isn't hung it is just taking a very long time. If you wait 10 years or so it may actually finish. The main point is that for a simulation this large you will need much more than just a dual processor machine to run it. Firstly a advise you to use PMEMD in place of sander.MPI, this will run much faster and require a lot less memory (as you increase the number of processors).

    You might also want to try setting ntpr=1 - this way you will be able to see that the minimization hasn't stopped at 500 steps but is in fact just running very slowly.

    Also a few comments on your input file:

    water minimization

     &cntrl

      imin=1, ntmin=1, nmropt=0,

      maxcyc=2000, ncyc=500, drms=0.1

      ntx=1, irest=0,

      ntpr=500, ntwr=500, iwrap=1,

      ntf=1, ntb=1, cut=8.0, nsnb=10,

      igb=0,

      ibelly=0, ntr=1,

      restraint_wt=300.0, restraintmask=':PEG'

     &end

    nsnb=10, this has no effect on its own, sander and pmemd will build the list automatically as needed.

    restraint_wt=300.0 - this is very large. It is probably okay for the minimization but not for when you do MD.

    You will also have to use the 'old' style of specifying restraints when you use pmemd.

    Good luck,

    Ross

    From: owner-amber_at_scripps.edu [mailto:owner-amber_at_scripps.edu] On Behalf Of Lili Peng
    Sent: Wednesday, December 17, 2008 8:36 AM
    To: amber_at_scripps.edu
    Subject: Re: AMBER: AMBER minimization problems

    Hi Ross,

    It just stops after the 2nd nstep and the minimization doesn't proceed any further. I'm attaching my minimization output for your reference.

    My input script was:
    /sander -O -i min_wat.in -o peg5k_wat_min_vac.out -p peg5k_wat.top -c peg5k_wat.crd -r peg5k_wat_min_vac.rst -ref peg5k_wat.crd

    My .crd and .top files are too large to attach to this email, so I've uploaded them to my web server here:
    http://bioengineering.ucsd.edu/~lpeng/peg/peg5k_wat.crd
    http://bioengineering.ucsd.edu/~lpeng/peg/peg5k_wat.top

    Thanks for your help,
    Lili

    2008/12/17 Ross Walker <ross_at_rosswalker.co.uk>

    Hi Lili,

    What do you mean by 'ends'? Do you get any further output? Any error messages of any kind? It would also be helpful to see your input file.

    All the best

    Ross

    From: owner-amber_at_scripps.edu [mailto:owner-amber_at_scripps.edu] On Behalf Of Lili Peng
    Sent: Tuesday, December 16, 2008 11:14 PM
    To: amber_at_scripps.edu
    Subject: AMBER: AMBER minimization problems

    Hi AMBER,

    I'm having trouble minimizing a 5000 Dalton polyethylene glycol molecule solvated in a TIP3PBOX 15 water molecules. The output file I get ends after the 2nd nstep of 500 (shown below). This is after I've vacuumed minimized my molecule (and then solvated it in water). I've attached my water minimization file input file. Do anyone know what the problem could be?

    Thanks,
    Lili

    --------------------------------------------------------------------------------
       4. RESULTS
    --------------------------------------------------------------------------------

     ---------------------------------------------------
     APPROXIMATING switch and d/dx switch using CUBIC SPLINE INTERPOLATION
     using 5000.0 points per unit in tabled values
     TESTING RELATIVE ERROR over r ranging from 0.0 to cutoff
    | CHECK switch(x): max rel err = 0.2738E-14 at 2.422500
    | CHECK d/dx switch(x): max rel err = 0.8314E-11 at 2.736960
     ---------------------------------------------------
    | Local SIZE OF NONBOND LIST = 64267381
    | TOTAL SIZE OF NONBOND LIST = 130524253

       NSTEP ENERGY RMS GMAX NAME NUMBER
          1 -2.0703E+06 1.2881E+01 4.6850E+01 O 255159

     BOND = 16.4676 ANGLE = 8.8005 DIHED = 158.2966
     VDWAALS = 255110.9185 EEL = -2325481.7410 HBOND = 0.0000
     1-4 VDW = 105.8433 1-4 EEL = -249.6019 RESTRAINT = 0.0000

       NSTEP ENERGY RMS GMAX NAME NUMBER
        500 -2.6528E+06 9.0587E-01 4.5299E+00 H1 309256

     BOND = 182580.9181 ANGLE = 4.6033 DIHED = 157.3146
     VDWAALS = 382897.9872 EEL = -3218340.8995 HBOND = 0.0000
     1-4 VDW = 103.2430 1-4 EEL = -249.2364 RESTRAINT = 3.5502
     EAMBER = -2652846.0697

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" (in the *body* of the email)
      to majordomo_at_scripps.edu