[gmx-users] Re: Gromacs on GPU

[Justin A. Lemkul]

Benjamin Hall wrote:
> Justin A. Lemkul wrote:
>> Ours is gcc-4.4.6 on a local machine using Cuda 3.2.  On our 
>> university's GPU 
>> cluster, the installation was done with gcc-4.3.4 and Cuda 3.1.  I 
>> have not 
>> tested Cuda 4.0, though it has recently become available to us so it 
>> might be 
>> worth a shot to see if this is another issue.
> 
> I've just noticed that issue 757 (initial temperature set to 0 for 
> mdrun-gpu) has also arisen in my simulations. Do you know if the source 
> of this error, and if there is a workaround for it?
> 

I don't know the source of it, no.  I would suspect that the interface between 
OpenMM and Gromacs doesn't properly handle constraints, velocities, T-coupling 
information, or a combination of both.  Implicit solvent calculations on CPU 
work completely as expected, so I suspect the problem lies outside of Gromacs. 
Somewhere along the way (in a bug discussion or on this list), I believe someone 
told me there were some issues related to OpenMM, but I'm not 100% sure on that. 
  It is also my understanding that the upcoming version of Gromacs is not going 
to be reliant on OpenMM.

In any case, the temperature eventually stabilizes to about the value set as 
ref_t, though the behavior described in bug #757 is present in anything I've done.

> Have you been able to run production simulations using mdrun-gpu?
> 

Not anything I've tried to publish, no.  I'm not totally confident in those 
results yet.  Running on CPU works fine, so it suggests to me that Gromacs is 
doing its job, but the external components are not.

-Justin

-- 
========================================

Justin A. Lemkul
Ph.D. Candidate
ICTAS Doctoral Scholar
MILES-IGERT Trainee
Department of Biochemistry
Virginia Tech
Blacksburg, VA
jalemkul[at]vt.edu | (540) 231-9080
http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin

========================================