[gmx-users] OpenMM Compatibility issues with GPU build

[Justin Lemkul]

On 12/19/13 2:02 AM, Seera Suryanarayana wrote:
> Dear Gromacs users,
>
> I have successfully installed
>
> a) cudatoolkit_4.2.9_linux_64_ubuntu11.04.run
> b) devdriver_4.2_linux_64_295.41.run
> c) gpucomputingsdk_4.2.9_linux.run
> d) OpenMM 4.1 from Source
> e) GROMACS 4.6.5
>
> While trying to install / cmake of mdrun-gpu using the standard procedure
> given in GROMACS site,
>
> export OPENMM_ROOT_DIR=path_to_custom_openmm_installation
>
> mkdir build_gromacs_gpu
> cd build_gromacs_gpu
> cmake .. -DGMX_OPENMM=ON [-DCMAKE_INSTALL_PREFIX=desired_install_path]
> make mdrun
> make install-mdrun
>
> I have the problem in the cmake step with following error message:
>
> -- Enabling native GPU acceleration
> CMake Error at src/contrib/CMakeLists.txt:49 (message):
>    The OpenMM build is not compatible with the native GPU build
>
>
> -- Configuring incomplete, errors occurred!
>
> Should I install a older version of OpenMM like 2.0 or is there any
> alternative method to build mdrun-gpu ?
>

OpenMM is no longer required in order to use GPUs.  The only feature (to my 
knowledge) that requires OpenMM is if you want to run an implicit solvent 
simulation on a GPU.  Even then, the supported features are minimal, hence why 
OpenMM support was moved to "contributed code" and will likely be dropped in the 
future.

-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

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