Using conda on the CQLS infrastructure

Our recommendation for new conda environments is to use the pixi tool for managing new conda environments. The majority of the installed software on the infrastructure is managed using pixi. pixi should be available to you if you have updated your dotfiles and enabled hpcman configuration.

There will be more information forthcoming about using pixi for reproducible research.

See here for info about using pixi in place of conda.

Using GPUs with pixi

To use GPUs with pixi, make sure to specify the appropriate version of cuda under the [system-requirements] entry in your pixi.toml. See here for more information

pixi.toml

[system-requirements]

cuda = "12.4"  # Replace "12.4" with the specific CUDA version you intend to use

Legacy conda installs

For legacy conda installs, we have updated our CQLS-managed conda install to use miniforge. Please run hpcman user setup conda to get this new conda install activated in bash.

Miniforge works similarly to the previously installed miniconda. The primary difference is that the defaults channel no longer points to repositories owned and managed by Anaconda. The defaults are now conda-forge and bioconda, and the packages in the base environment are from conda-forge.

What changed?

Anaconda has recently updated their licensing terms such that packages in the defaults channel are a paid service for organizations with more than 200 employees. See more about Anaconda pricing here.

The CQLS does not and will not pay for an Anaconda license at this time, and does not support installing packages from the defaults channel.

All currently installed and managed conda environments are configured to use only the conda-forge and/or bioconda channels.