LAMMPS Windows Installer Repository

This repository is hosting pre-compiled Windows installers of the LAMMPS molecular dynamics simulation software package. The binaries are built with MinGW64 Linux to Windows cross compilers using the standard LAMMPS sources. The LAMMPS binaries contain all optional packages included in the source distribution except: KIM (license is not GPL compatible), KOKKOS and USER-INTEL (do not support cross-compilation with GCC), MSCG (requires an external library), USER-H5MD (requires an external library) PYTHON (requires to bundle a full Python runtime), USER-NETCDF (requires an external library), USER-QMMM (only useful when linking to a QM software), USER-QUIP (requires an external library), USER-VTK (requires an external library), REAX (superseded by the USER-REAXC package which is included). The serial executable additionally does not contain the MPIIO and USER-LB packages, since those require MPI-IO functions, which are not available without linking to a real MPI library.

Some Notes on GPU Support
These Windows binaries include (experimental on Windows) GPU acceleration via the GPU package. This is achieved through compiling the GPU package in OpenCL mode and linking to an OpenCL v1.2 compatible ICD loader. This means the executables do not contain any vendor provided code and should be compatible with GPUs from both AMD and Nvidia. The GPU package has been compiled for mixed precision computation and is currently somewhat tuned for Nvidia (Kepler generation) GPUs. It does not yet work with OpenCL drivers for CPUs (like those included in the Intel and AMD OpenCL SDKs).
In the case of having multiple Vendor provided OpenCL runtimes installed, you may run into the situation of the "wrong" runtime being set as the default and used by LAMMPS. In this case, you may need to remove unwanted vendors from the windows registry database. Look for the key: HKEY_LOCAL_MACHINE\SOFTWARE\Khronos\OpenCL\Vendors, and remove vendors such as Intel. When reporting problems, please always include the exact version of the installer and the output of the ocl_get_devices tool.

Looking for pre-compiled Linux binary RPMs? They are just a mouse click away.

Installing LAMMPS on Windows

There are installer packages for 32-bit and 64-bit versions of Windows available. The respective download directory will contain installer packages that are labeled with the date of the LAMMPS version and packages labeled as latest. It is usually recommended to download and install the latest package. The other packages are provided in case there is a problem with it. Download the installer executable suitable for your machine, execute it, and follow the instructions in the dialogs. Each version will install into a different directory, so it is possible to have multiple versions installed at the same time (however it is not recommended). Both kinds of packages contain:

Both executables will run in serial when executed directly. Please see below for instructions on how to perform parallel runs. To use the MPI based parallelism, you also need to install MPICH2 from Argonne lab. For 32-bit Windows you have to download and install mpich2-1.4.1p1-win-ia32.msi or any compatible version. Correspondingly, for 64-bit Windows you have to download and install mpich2-1.4.1p1-win-x86-64.msi or any compatible version.

Running LAMMPS on Windows

General Comments

LAMMPS is a text mode program, so you have to click on the Command Prompt entry in the windows start menu, launch the Windows Power Shell, or simply execute the command cmd from Start -> Run. The LAMMPS manual and the colvars reference manual in PDF format are linked from the start men, also references to open the file system browser to view the Benchmarks and Examples directory tree. The PATH environment variable is automatically set by the installer, so that the bundled executables from everywhere in the system. Also the LAMMPS_POTENTIALS environment variable is preset to the folder with the bundled manybody potential files, so they do not need to be copied into the respective working directory.

Serial Execution

To run a LAMMPS simulations open a terminal window as outlined above and change to the directory with your input script and then run either lmp_serial or lmp_mpi As described in the LAMMPS manual. For more details and additional command line flags, please check out the LAMMPS Manual.

Multi-threaded Parallel Execution

All LAMMPS binaries from this repository support multi-threading via OpenMP, however by default only one thread is enabled. To enable more threads, e.g. four, you need to either set it at the command line prompt via set OMP_NUM_THREADS=4 or use the package omp command in your input script.

After enabling the use of multiple threads, you also have to select styles in LAMMPS, that are multi-thread enabled. Those are usually identified by the /omp suffix. Those can be selected explicitly and selectively, or with the suffix command or - most conveniently - via using the -sfs omp command line flag to the LAMMPS executable.

Message Passing (MPI) Parallel Execution

Only the lmp_mpi executable supports parallel execution via MPI (which can be combined with OpenMPI multi-threading). For that you also need to install MPICH2 from Argonne lab. The installer does not contain it and does not check for it.

After the installation of the MPICH software, it needs to be integrated into the system. For this you need to start a Command Prompt in Administrator Mode (right click on the icon and select it). Change into the MPICH2 installation directory, then into the subdirectory bin and execute smpd.exe -install.

MPI parallel executables have to be launched with mpiexec -localonly # lmp_mpi ... or mpiexec -np # lmp_mpi ... with "#" being then number of parallel processes to start. For the latter you may be asked for your password.

Known Issues: while running in MPI parallel mode any output to the screen is block buffered and not line buffered as usual. That is, output will not show up immediately, but only after a certain amount of output has accumulated. This is done for efficiency reasons. Multi-threaded parallel or serial runs are not affected. Also when running under mpiexec, you should not use multi-threading since MPICH restricts processes to a single physical CPU thus it will run less efficient.

For more details, please refer to the MPICH2 Documentation.
Axel Kohlmeyer
Last modified: Thu Nov 23 16:46:56 EST 2015