We ise Coder at our lab for machine learning simulations. We have a self hosted Coder deployment and use docker to proviosn workspaces on our GPU servers. Each lab member can spin up a fully working, dev environment in less than a minute and they don't ever have to think about installing it configuring the. Correct CUDA version, GPU drivers etc.

We persist home directory of a user across all their workspaces and also mount a common data directory to all so that we have access to ore downloaded datasets.