Glotzerlab software

glotzerlab-software deploys software developed by the Glotzer group at the University of Michigan to HPC resources via conda packages. Use glotzerlab-software to install an MPI and GPU enabled build of HOOMD-blue and related packages on a HPC resource.

Use the standard conda-forge provided HOOMD-blue (mamba install hoomd with no special configuration) for serial execution on individual workstations.

Resources

Quick start

  1. Install miniforge.

  2. Replace $HOME/miniforge3/.condarc with:

    channel_priority: strict
    channels:
      - file://path/to/conda/channel
      - conda-forge
    
    disallow:
    - openmpi
    - mpich
    - cuda-cudart-dev
    
  3. Install HOOMD-blue:

    mamba install hoomd
    
  4. Load prerequisite modules. For example:

    module load openmpi cuda
    
  5. Execute Python scripts with MPI and GPU support. For example:

    srun -n 8 python3 script.py
    

You can confirm that the correct package was installed with mamba list hoomd. The full package build will contain the cluster name and a gpu or cpu suffix. For example:

# Name      Version   Build                          Channel
hoomd       X.Y.Z     cluster_gpu_py311h1b32822_0    file://path/to/conda/channel

Contributing

Contributions are welcomed via pull requests. Please report bugs and suggest feature enhancements via the issue tracker. See ARCHITECTURE.md for information on how the repository is structured, including how to modify the containers.


Development of glotzerlab-software is led by the Glotzer Group at the University of Michigan.

Copyright © 2024 The Regents of the University of Michigan.