[gmx-users] GPU g_energy problem

[Justin Lemkul]

On Mon, Nov 25, 2013 at 3:40 AM, Andrei Neamtu <neamtuandrei at gmail.com>wrote:

> Dear GMX users,
>
> I am simulating a system of two proteins in TIP4P water box. The
> simulations are performed on a system with 3 servers each containing 16 CPU
> cores and 2 Tesla 2090 GPU.
> The problem thai I encounter is related to the saved interaction energies
> in the .edr file and read with g_energy.
>
> It seems that when using the GPU acceleration none of the interaction
> energies are saved for the short-range non-bonded interaction computed on
> the GPU.
> The SOL (solvent) - SOL LJ-SR and Coul-SR are all ZERO. Also Protein - SOL
> are all zero.
> When using CPU only accceleration these energies are saved correctly!
> Despite the zeroes in the energy file the simulations seems to proceed
> correctly when visually inspected.
>
> I wonder in I do not miss something obvious.
>
> I am asking if anybody observed a similar behaviour and what should be the
> solution to this problem.
>
>
> I upgraded to 4.6.4 version but also on 4.6.1 the problem was still
> present.
>
>
Can you clarify if you have re-run the simulation under 4.6.4, not just
re-analyzed the energies with 4.6.4 g_energy?  Doing the latter will have
no effect if the energies were not saved correctly in the original .edr
file.  http://redmine.gromacs.org/issues/1293 seems relevant here; perhaps
multiple energygrps still are not supported on GPU.

-Justin

-- 

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

Justin A. Lemkul, Ph.D.
Postdoctoral Fellow

Department of Pharmaceutical Sciences
School of Pharmacy
Health Sciences Facility II, Room 601
University of Maryland, Baltimore
20 Penn St.
Baltimore, MD 21201

jalemkul at outerbanks.umaryland.edu | (410) 706-7441


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