[AMBER] pmemd.cuda unstable with ff19SB+OPC and HMassRepartition

From: Tucker Burgin <tburgin.uw.edu>
Date: Wed, 10 Mar 2021 14:25:22 -0500

Hi Amber users,

I'm trying to develop a workflow for high-throughput simulations of
proteins. I've got an enzyme-substrate complex parameterized with ff19SB in
a box of OPC waters, to which I've applied hydrogen mass repartitioning
using ParmEd with the default settings. I minimized, heated from 100 to 300
K, and equilibrated for 1 ns using pmemd.MPI, and all that is very
well-behaved.

Now I'm trying to run longer simulations with the equilibrated structure
using pmemd.cuda, but whenever I try to run a simulation with a timestep
longer than 2 fs, it runs for a while (anywhere from a few hundred ps to
several ns) before crashing with a generic error message:

cudaMemcpy GpuBuffer::Download failed an illegal memory access was
encountered

This happens very consistently. I do *not *have this problem if I reduce
the timestep to 2 fs, but I don't understand why. I have confirmed that the
hydrogen mass repartitioning is correctly reflected in the topology file,
and SHAKE is turned on. I have tried manually restarting the crashed
simulation many times thinking that it maybe just needed more
equilibration, but after several restarts I'm seeing no improvement. Here's
my input file:

 &cntrl

  imin=0,

  ioutfm=1,

  ntx=1,

  ntxo=1,

  nstlim=2500000,

  dt=0.004,

  ntf=2,

  ntc=2,

  tempi=300.0,

  temp0=300.0,

  ntpr=25000,

  ntwx=25000,

  cut=8.0,

  ntb=2,

  ntp=1,

  ntt=2,

  vrand=10000,

  ig=-1,

  taup=2,

 &end

As a further test, I have tried removing the substrate from the model
altogether, leaving only a protein consisting entirely of standard
residues, a single sodium ion, and the OPC water molecules, and still the
problem persists.

The output files look fine, with reasonable values for the temperature, all
the energy terms, and the density. The only indication I have of anything
weird in the output files is that the temperature at the 0th step is always
about 436 K (even though tempi is 300 K and the system has been
equilibrated at that temperature), but it returns to about 300 K by the
next write to the output file. Here's an example output line from step 0:

 NSTEP = 0 TIME(PS) = 0.000 TEMP(K) = 436.04 PRESS =
-19.0

 Etot = -151011.7523 EKtot = 47799.7254 EPtot =
-198811.4778

 BOND = 1683.1249 ANGLE = 4073.2499 DIHED =
2465.2522

 UB = 0.0000 IMP = 0.0000 CMAP =
381.8329

 1-4 NB = 1689.6414 1-4 EEL = 19887.1333 VDWAALS =
21224.1537

 EELEC = -250215.8661 EHBOND = 0.0000 RESTRAINT =
0.0000

 EKCMT = 13861.9920 VIRIAL = 14078.6721 VOLUME =
527955.5220

                                                    Density =
1.0336

There's nothing obviously weird to my eye going on in the simulation
trajectory. I'm using Amber 20. Since I can't find anything wrong with the
simulations I'm considering just writing a script to automatically restart
my simulations whenever they crash, but I'd feel a lot more comfortable if
I knew for sure what the problem was. Anyone have any ideas or suggestions?
Please let me know if there's any more information I can provide.

Thanks!
-Tucker Burgin
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Mar 10 2021 - 11:30:02 PST
Custom Search