On Sat, Apr 24, 2010, Levi Pierce wrote:
>
> I have been running pmemd.cuda on several different
> small peptide systems. After minimization and equilibration
> with sander I run md with pmemd.cuda which produces the
> following output in the mdinfo file:
>
> NSTEP = 1000 TIME(PS) = 419.000 TEMP(K) = 312.76 PRESS =
> 0.0
> Etot = ************** EKtot = 300.8112 EPtot =
> **************
>
> Here is the input file I used for running the tests:
>
> CYC MD
> &cntrl
> imin=0,irest=1,ntx=5,
> nstlim=25000000,dt=0.002,ntb=0,
> tempi=300, temp0=300,
> ntf=2,ntc=2,tol=0.000001,
> ntpr=1000, ntwx=1000, ntwr=1000,
> cut=9999.0, rgbmax=15.0,gbsa=0,
> igb=5,ntt=3,nscm=1000,gamma_ln = 1.0
> /
Just to add to what Ross said, (and for everyone's benefit): the first
thing to do when something like this arises is to try runs with ntpr=1 and
a small number of steps. That is, use nstlim=50 instead of 25000000(!)
and ntpr=1 instead of 1000.
Also, but of less importance, is to make tol an order of magnitude smaller,
and try dt=0.001 rather than 0.002.
Such tests will allow one to see if the initial energies and forces are the
same as on a cpu or not, and sometimes allow problems to be diagnosed that are
hidden when you wait 1000 steps before printing out the first results.
....regards...dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sun Apr 25 2010 - 09:30:03 PDT
