Hi Mani,
The hydrogen mass is hardwired in pmemd.cuda in AMBER 12 so if you are
modifying it in the prmtop file the sampling will be way off. If you have
AMBER 14 you should use that since that supports varying masses for
hydrogen.
If you only have AMBER 12 then first make sure you are fully up to date on
all patches - the mdout files should be reporting v12.3.1. Then edit
gputypes.cpp in $AMBERHOME/src/pmemd/src/cuda/ and change line 33 from
1.008 to the mass that your hydrogens should be. Then recompile
pmemd.cuda. (remember that you did this though since it will mess up
normal calculations).
If you want mixed masses for hydrogen then you will need to update to
AMBER 14.
Note with masses scaled to half their value it's possible that the system
won't be stable at 2.0fs. It is probably on the edge of stability as is
and shake is holding you together. The GPU shake 'may' be less tolerant of
integration errors than the CPU version so you may still have issues that
require the timestep to be reduced.
All the best
Ross
On 6/25/14, 2:56 PM, "Manikanthan Bhavaraju"
<manikanthanbhavaraju.gmail.com> wrote:
>Hi,
>
>I am observing sampling issues with pmemd.cuda of Amber12 for a monomer
>system. Currently, I am using ff9SB forcefield and an explicit solvent
>model where the masses of solvent and side chains are rescaled by a factor
>of 0.5. I have tested the performance of pmemd.MPI and pmemd.cuda. The
>system is stable with pmemd.MPI version. The monomer was minimized and
>sampled under NVT conditions for 50 ns using the following input files:
>
>Minimization step1:
>
>&cntrl
>
> imin = 1,
>
> maxcyc = 5000,
>
> ncyc = 2500,
>
> ntb = 1,
>
> ntr = 1,
>
> cut = 12.0,
>
> /
>
>Hold the protein weakly
>
>10.0
>
>RES 1 107
>
>END
>
>END
>
>Minimization Step2
>
>1REI: initial minimization of the system
>
> &cntrl
>
> imin = 1,
>
> maxcyc = 5000,
>
> ncyc = 2500,
>
> ntb = 1,
>
> ntr = 0,
>
> cut = 12.0
>
> /
>
>Generating initial velocities under NVT conditions:
>
>&cntrl
>
> imin = 0,
>
> irest = 0,
>
> ntx = 1,
>
> ntb = 1,
>
> cut = 12.0,
>
> ntr = 1,
>
> ntc = 2,
>
> ntf = 2,
>
> ntt = 3,
>
> gamma_ln = 1.0,
>
> nstlim = 50000,
>
> tempi = 298.15,
>
> temp0 = 298.15,
>
> dt = 0.002,
>
> ntpr = 1000,
>
> ntwx = 1000,
>
> ntwr = 1000,
>
> /
>
>keep the protein fixed with weak restraints
>
>10.0
>
>RES 1 107
>
>END
>
>END
>
>Sampling process:
>
>&cntrl
>
> imin=0,
>
> ntx=5,
>
> irest=1,
>
> ntc=2,
>
> ntf=2,
>
> nstlim=5000000,
>
> ntt=3,
>
> gamma_ln=1.0,
>
> temp0=298.15,
>
> ntpr=5000,
>
> ntwr=5000,
>
> ntwx=5000,
>
> dt=0.002,
>
> ig=-1,
>
> ntb=1,
>
> cut=12.0,
>
> ioutfm=1,
>
> /
>
>
>
>The average RMSd over 50ns was 1.07 +/- 0.01 for the monomer when
>pmemd.MPI
>was used. The structures at various intervals were visualized using VMD.
>The 3D structure of the monomer was retained throughout the simulation.
>
>Then, I have used pmemd.cuda in order to run the system for a longer
>simulation time. The generation of initial velocities and sampling were
>done using a similar input file that was used for pmemd.MPI. However, the
>RMSd of the monomer just after 10 ps of the simulation was about 7.0
>angstrom and keeps increasing all the way to 36.0 angstroms after 10 ns of
>simulation. The tertiary structure of the monomer was lost and the
>protein
>has denatured. So, I have tested the behavior of the system using 1 fs
>time step. The system was stable until 10 ns of the simulation (RMSd 1.5
>+/- 0.3). But the extended simulation has once again denatured the
>protein and the RMSd creeps up quickly.
>
>I don't understand why the system is getting destabilized with cuda ? Can
>someone comment on this issue?
>
>Thanks,
>
>Mani
>
>
>
>--
>Manikanthan Bhavaraju
>_______________________________________________
>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 Jun 25 2014 - 18:00:02 PDT
