From: Francesco Pietra (francesco.pietra_at_accademialucchese.it)
Date: Fri Apr 11 2014 - 11:39:10 CDT
accelMD on GPUs was not advertised, as far as I know. If it was activated
in the nightly builds of either 2.9 or 2.10 would be great. We are loosing
a lot of time by using CPUs only. Hope developers would like to confirm. I
asked some days ago about a rough date for namd being able to compute
energies on GPUs, having no answer.
On Fri, Apr 11, 2014 at 2:19 PM, Jeff Wereszczynski <jwereszc_at_iit.edu>wrote:
> Hi Kevin,
> Which version of NAMD are you using? I don't think aMD was working on the
> GPUs for the nonbonded terms in version 2.9, but it looks like they've
> fixed it in the nightly builds (I haven't tried it though).
> Also, are you turning on your aMD forces at 300K? You may want to try
> heating your system with the aMD forces on, or even minimizing then heating
> your system with the aMD forces on.
> Jeff Wereszczynski
> Assistant Professor of Physics
> Illinois Institute of Technology
> On Fri, Apr 11, 2014 at 3:27 AM, Felipe Merino <
> felipe.merino_at_mpi-muenster.mpg.de> wrote:
>> The rules for defining the acceleration parameters that you cite were
>> meant to be used with proteins primarily as far as i remember. Actually,
>> they are more a rule of thumb than anything else. For systems with lipids
>> the choice can a bit different as the dihedral energy per residue on a
>> lipid is quite different than an aminoacid.
>> Take a look at this paper, the define the parameters based directly on
>> the dihedral energy Miao Y, Nichols SE, McCammon JA (2014). Physical
>> Chemistry Chemical Physics 16(14):6398- 6406.
>> Have in mind that for big systems the acceleration (dV) gets a bit too
>> big and then in practice you cannot reweight the final distribution simply
>> because you average gets dominated by conformations that you rarely sample
>> (very high exp(dV*beta)
>> Being that said, I never saw my aMD simulations fail with a constraint
>> failure even with pretty huge acceleration (I even use it sometimes to
>> remove gross error from initial models) which
>> suggests that your systems have issues unrelated to the acceleration.
>> Hope that helps
>> On 04/10/2014 08:40 PM, Kevin Kastner wrote:
>> Hello all! I am trying to use accelerated molecular dynamics on NAMD
>> for a protein that is inside of a lipid bilayer and am having trouble.
>> Previously, I had set up and run the simulation using the NAMD membrane
>> tutorial and have gotten that to work (though I have converted it to use
>> the amber force field instead of CHARMM). From other sources I have
>> calculated the energies and alphas for each of the protein + system and the
>> membrane to be the following:
>> accelMDE = <dihedral energy from standard simulation> + 4 * (# of
>> residues) + 40 * (# of POPC molecules)
>> accelMDalpha = 1/5 * 4 * (# of residues) + 20 * (# of POPC molecules)
>> accelMDTE = <total energy from standard simulation> - <dihedral energy
>> from standard simulation> + 0.16 * (# of total atoms)
>> accelMDTalpha = 0.16 * (# of total atoms)
>> I have tried running with these new parameters both starting after
>> the three preparation steps that need to be done with transmembrane systems
>> (melting the lipid tails, then constraining the protein, and finally
>> letting the whole system equilibrate) and also starting at the very
>> beginning with the three steps using aMD, but the system crashes after a
>> short while every time. Usually the error has to do with the SHAKE or
>> RATTLE algorithms.
>> Does anyone know anything else I need to do to get this to work? Do I
>> need to reduce the size of the timesteps from 2fs to 1fs? Or do I need to
>> do something else? Oh, I do not know if this makes a difference, but I am
>> also trying to run this on GPUs, rather than CPUs. Is aMD not compatible
>> with GPU-accelerated MD? I would appreciate any help that anyone can
>> provide. Thank you!
>> - Kevin Kastner
>> “For me, insanity is super sanity. The normal is psychotic. Normal means
>> lack of imagination, lack of creativity.” - Jean Dubuffet
This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:22:19 CST