FAHBench is the official Folding@Home benchmark. Like the Folding@Home "cores" being executed by hundres of thousands of donors across the world to solve hard problems in protein dynamics, FAHBench is built on the molecular dynamics engine OpenMM. FAHBench works on any OpenCL device, as well as CUDA-capable Nvidia GPUs. FAHBench is available as a GUI or command-line based application for Windows and Linux. It is licensed under GPLv2.

The benchmark runs a short molecular dynamics (MD) simulation on various systems. The score is the number of MD steps completed per unit time.

The score depends strongly on the size of the system being simulated. Traditionally, explicit solvent DHFR (~20k atoms) is the benchmarking system of choice. FAHBench includes other systems ("workunits") for benchmarking. A scaled score is reported to compare results on different-sized systems to DHFR. Please take this scaled score with a grain of salt.

The score is quantified with the peculiar unit of "nanoseconds per day". This gives the ratio of simulated time to real-life time.

The theoretical scaling of molecular dynamics is N log N (where N is the number of atoms). This factor is used to provide a "scaled score" comparable to a DHFR-sized system. In practice, scaling is rarely N log N.

OpenMM supports CPU calculations as well. This capability is not included in the released binaries. You can compile OpenMM with the CPU platform enabled and test it with the --platform CPU argument on the command line. Please know that CPU calculations on Folding@Home are performed with GROMACS, a more performant CPU code.

The parameters of a simulation are specified with three files from a standard Folding@Home work unit (WU):

• system.xml
• integrator.xml
• state.xml

as well as a FAHBench-specific file named wu.json. This file encodes the number of steps to perform and other meta-information. To install a custom work unit, copy these four files into a subdirectory of share/fahbench/workunits/.

FAHBench is built with CMake and requires the following libraries:

• Boost - automatically downloaded and built
• OpenMM (source) - molecular dynamics calculations.
• OpenCL - linked to find OpenCL devices.
• CUDA runtime - optional - linked to find CUDA devices.
• Qt5 - optional - for building the GUI.

1. Get the prerequisites

   sudo apt-get install \
       qt5-default \
       nvidia-cuda-dev nvidia-opencl-dev # are there generic packages for opencl?

2. Configure an OpenMM build with CMake.

3. From a clean build directory

   ccmake [source_dir]
   # make sure all dependencies are found
   # type c - configure, g - generate
   make
   make install

1. Download and install

   • Visual Studio Community 2013
   • Qt > 5.2
   • CMake > 2.8.11
   • AMD APP SDK
   • (Optional) CUDA SDK = 6.5

   These all have nice GUI installers.

2. Download OpenMM and build with CMake. I don't think the provided binaries (VC2010) will work. Build the release configuration. You can disable building the python bindings (which may be a source of build errors). Build the INSTALL project to install OpenMM.

3. Run CMake on the fahbench source directory. Finagle it until it has found all of the dependencies you just spent so long getting in order. Start by setting:

   • CMAKE_PREFIX_PATH to .../Qt/5.4/msvc2013_64/
   • OPENMM_xxx to where you installed OpenMM.

4. Build and install! CMake will copy the relevant OpenMM and Qt dlls to the bin/ install directory.

1. Make sure your git submodules are initialized (openmm)

2. Download a .tar.bz2 release of the AMDAPPSDK into this directory. The provisioning script can't download it because you have to accept a license agreement through a web browser. The file has to be of the form AMD-APP-SDK-*.tar.bz2

3. Run vagrant up.

4. Enter the virtual machine with vagrant ssh

5. Run install-openmm.sh, then install-fahbench.sh. Build artifacts will be copied to the dist/ directory in this repository.