Installing binaries¶

MPI parallel builds¶

You must build HOOMD-blue from source to enable support for the native MPI and CUDA libraries on your HPC resource. You can use the glotzerlab-software repository to manage such builds as conda packages.

Serial CPU and single GPU builds¶

HOOMD-blue binaries for serial CPU and single GPU are available on conda-forge for the linux-64, osx-64, and osx-arm64 platforms. Install the hoomd package from the conda-forge channel:

Pixi

pixi add hoomd=7.0.1

Micromamba

micromamba install hoomd=7.0.1

Mamba

mamba install hoomd=7.0.1

By default, micromamba 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:

Pixi

First add the following to your pixi.toml file:

[system-requirements]
cuda = "12.9"

Then, add hoomd with:

pixi add "hoomd=7.0.1=*gpu*"

Micromamba

export CONDA_OVERRIDE_CUDA="12.9"
micromamba install "hoomd=7.0.1=*gpu*" "cuda-version=12.9"

Mamba

export CONDA_OVERRIDE_CUDA="12.9"
mamba install "hoomd=7.0.1=*gpu*" "cuda-version=12.9"

Note

conda-forge may update to a new version of CUDA after these instructions are published. If the above command results in an error, replace 12.9 with the version noted in micromamba’s error message.

Similarly, you can force CPU-only package installation with:

Pixi

pixi add "hoomd=7.0.1=*cpu*"

Micromamba

micromamba install "hoomd=7.0.1=*cpu*"

Mamba

mamba install "hoomd=7.0.1=*cpu*"