On Mon, 2014-04-28 at 18:05 +0200, Pierpaolo Cacciotto wrote:
> Dear all,
>
> I would like to use the new Amber14 pmemd.cuda to run a simulation of a
> protein in explicit solvent (trucated octahedral box, ~180,000 atoms)
> using a GeForce GTX 780. I used ParmEd to edit the old toplogy file (the
> one I am still using with Amber12) and I created a new one with H-masses
> set to 4.032 in order to use the 4fs timestep. I used the new topology and
Try the default Hydrogen mass to start with (3.024). The mass of 4.032
may be a little too high. For methyl groups, the mass of the carbon
atom will be reduced to ~2-3 amu. At this point you have repartitioned
_too_ much.
> a new rst file (generated with ParmEd) to run the simulation. It runs for
> a while but then stops with the following message
>
> "Error: unspecified launch failure launching kernel kClearForces
> cudaFree GpuBuffer::Deallocate failed unspecified launch failure"
>
> Could you suggest me how to fix this error?
Error messages from the GPU are typically not helpful. It's the
equivalent of saying "something went wrong" and so are impossible to
diagnose directly. Do you see the problem on the CPU? Try to "catch"
the program in the act of destroying the system (print very often and
watch the trajectory). You can run the simulation up until right before
it crashes and then write a restart file (use ntxo so you get full
double precision restarts). Then hopefully you can run simulations
setting ntpr=1 and ntwx=1 so you can see what happens when the
calculation dies by watching each step.
If the problem is only found on the GPU, try compiling with DPFP instead
and see if it is related to the precision model. You're kind of pushing
the edge of stability here with a technique not many people have tried
with Amber. So it's not surprising some debugging effort is necessary.
HTH,
Jason
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Apr 28 2014 - 10:00:03 PDT
