Currently, Crusher and Frontier do NOT have Anaconda/Conda modules. If your workflow better suits conda environments, you can install your own Miniconda on Frontier.
The install process is rather simple (with a few notable warnings, see Cautionary Notes further below):
mkdir miniconda_frontier/ cd miniconda_frontier/ wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh chmod u+x Miniconda3-latest-Linux-x86_64.sh ./Miniconda3-latest-Linux-x86_64.sh -u -p ~/miniconda_frontier
-pflag specifies the prefix path for where to install miniconda.
-uupdates any current installations at the
-plocation (not necessary if you didn’t do a “mkdir” beforehand)
While running the installer, you will be prompted with something like this:
Do you wish the installer to initialize Miniconda3 by running conda init? [yes|no]
If “yes”, your
.bashrc (or equivalent shell configuration file) will be updated with something like this:
# >>> conda initialize >>> # !! Contents within this block are managed by 'conda init’ !! . . . . . #unset __conda_setup # <<< conda initialize <<<
By default, this will always initialize conda upon login, which clashes with other Python installations (e.g., if you use the anaconda modules on other OLCF systems).
It is MUCH SAFER to say “no” and to just export the
PATH manually when on Frontier to avoid clashing:
.bashrc already has a similar block of code (e.g., from other OLCF modules), then it will NOT modify your
An additional recommendation is to set things to not activate your base environment by default (to help with the potential clashing):
# Only needs to be run once after exporting conda into your PATH conda config --set auto_activate_base false
As always, if you encounter issue, don’t hesitate to contact email@example.com