Installing binaries#

HOOMD-blue binaries are available in the glotzerlab-software Docker/Singularity images and in packages on conda-forge

Singularity / Docker images#

See the glotzerlab-software documentation for instructions to install and use the containers on supported HPC clusters.

Conda package#

HOOMD-blue is available on conda-forge on the linux-64, osx-64, and osx-arm64 platforms. Install the hoomd package from the conda-forge channel into a conda environment:

$ conda install hoomd=4.5.0

conda auto-detects whether your system has a GPU and attempts to install the appropriate package. Override this and force the GPU enabled package installation with:

$ export CONDA_OVERRIDE_CUDA="12.0"
$ conda install "hoomd=4.5.0=*gpu*" "cuda-version=12.0"

Similarly, you can force CPU only package installation with:

$ conda install "hoomd=4.5.0=*cpu*"

Note

CUDA 11.2 compatible packages are also available. Replace “12.0” with “11.2” above when installing HOOMD-blue on systems with CUDA 11 compatible drivers.

Note

To use Run time compilation on macOS, install the compilers package:

$ conda install compilers

Without this package you will get file not found errors when HOOMD-blue performs the run time compilation.

Tip

Use mambaforge, miniforge or miniconda instead of the full Anaconda distribution to avoid package conflicts with conda-forge packages. When using miniconda, follow the instructions provided in the conda-forge documentation to configure the channel selection so that all packages are installed from the conda-forge channel.