2026-11-04 · belial and fierna maintenance Saturday night. The nodes Fierna and Belial are short on ephemeral storage, so I need to reconfigure some paths to the local raid. Unfortunately, this requires draining the nodes of all pods first, so computations will be interrupted. Pods under a controller (e.g. a Job) will restart after the maintenance windows is over. Expected downtime is 2h per node, as I will use this opportunity for system upgrades.
2026-11-03 · kubeconfig update, fewer logins. If kubectl sends you to the browser login every day, your kubeconfig is from before this date: download kubeconfig-ccu.yaml again (or add `--oidc-extra-scope=offline_access` to its args) and log in once. Afterwards the login is renewed silently and only comes back after 30 days without use. More
2026-10-22 · New documentation site. The cluster documentation moved from the wiki to this site. The quick start describes the current login (kubectl oidc-login). Feedback to ccu@uni-konstanz.de. More
2026-10-20 · Christmas maintenance window. The cluster will be down for updates between Christmas and New Year. Exact dates follow by mail; plan long training runs accordingly and keep checkpoints.
A Kubernetes cluster with GPUs for research groups of the university. You describe a container to run, the cluster finds a machine with a free GPU for it. Accounts, questions and problems: ccu@uni-konstanz.de.
New here?
Install two tools, log in, run your first GPU job. About half an hour.
Quick start →Documentation
GPUs, storage, training jobs, container images, troubleshooting.
Read the docs →Cluster status
Free GPUs per node right now, outages, storage, open alerts.
Open the status page →Login
- Logging in with kubectl the normal way: browser login through the kubectl plugin
- Login page a kubeconfig with a token, without the plugin
- Password reset also for activating a new account
- Change password
Services
- Container registry browse images; push to
ccu-k8s.inf.uni-konstanz.de:32250 - Grafana all dashboards, including GPU utilisation per node
- Wiki user-written notes and recipes (login required)
- JupyterHub for lectures, separate login
Help
- Troubleshooting pending pods, OOMKilled, ImagePullBackOff, Unauthorized
- Rules how to be a good neighbour on a shared cluster
- News maintenance windows and changes
- ccu@uni-konstanz.de everything else