CPU & GPU Quota
UZH JupyterHub tracks how long you use a server each week. Different hardware tiers — CPU and GPU — have separate time budgets. When your weekly budget runs out, your server is stopped and you cannot start a new one until the next calendar week.
How the quota system works
Every server profile belongs to a tier — either CPU or GPU (determined by the profile name). Each user also has a role — user, admin, or collab — derived automatically from your OLAT membership (see Roles, Groups & Collaboration).
The combination of tier and role determines your weekly time budget:
| Tier | Role | Typical default budget |
|---|---|---|
| CPU | user | 20 hours / week |
| GPU | user | 20 hours / week |
| CPU | admin | 20 hours / week |
| GPU | admin | 20 hours / week |
| CPU | collab | 20 hours / week |
| GPU | collab | 20 hours / week |
Actual values are configured per deployment and may differ from the defaults above.
The quota resets every Monday at 00:00 UTC. Usage from the previous week does not carry over.
What happens when you run out
- Your running server is automatically stopped when the remaining time reaches zero.
- Attempting to start a new server returns a 403 error with a message showing how much time you used.
- The budget resets at the start of the next calendar week.
Seeing your remaining time
On the server-selection page
When you log in and choose a server profile, a Quota status banner shows your remaining time for each available profile:
| Profile | Tier | Remaining | Total |
|---|---|---|---|
| Minimal environment | CPU | 18h 30m | 20h 0m |
| BME GPU Server | GPU | 20h 0m | 20h 0m |
Inside JupyterLab — profile switcher
Once your server is running, the Switcher extension (a JupyterLab toolbar button) allows you to:
- View your remaining quota for the current profile
- Switch to another profile (if you have quota left on it) without stopping your server
The switcher displays an estimate of remaining time for the current and, as a mouse-over hover, other profile(s) - see the image below. A countdown timer runs in the background; your server is automatically stopped when time runs out.
What to do when your quota is used up
👩🎓 Participants:
- You cannot extend your own quota. If you run out, wait until Monday or ask your teacher to grant extra time.
- To make your budget last: stop your server when you are not actively working (use the Stop My Server button on the JupyterHub home page or the Start & Scale UI).
- Close idle notebook kernels — a running kernel keeps the server alive and consuming quota.
👩🏫 Owners / Coaches:
You can grant extra time to individual users via the Quota Admin UI (see below).
Quota Admin UI
👩🏫 Owners / Coaches:
The quota system includes a built-in admin panel at <your-hub-url>/hub/services/quota/admin. Access requires the admin scope (granted automatically to OLAT course owners).
Viewing usage
The admin page shows a table of all users who have used quota in the current week:
| Username | Tier | Role | Used | Budget |
|---|---|---|---|---|
| alice | cpu | user | 12h 30m | 20h 0m |
| bob | gpu | user | 3h 15m | 20h 0m |
You can filter by username using the search field at the top.
Extending a user’s quota
To grant extra time to a user who has run out:
- Find the user in the admin table.
- Enter the number of additional hours in the Extend field.
- Click Extend.
This subtracts the granted hours from the user’s recorded usage, effectively giving them more time within the current week.
Adjusting budgets
At the bottom of the admin page, a budget table lets you change the weekly time budget per tier and role. You can set separate values for:
- This week — takes effect immediately for the current period.
- Next week — takes effect from the following Monday.
This allows you to, for example, increase the GPU budget for a week when students have a deadline, then let it revert to normal.
How quota is configured (deployment level)
👩🏫 Owners / Coaches:
Default budgets
Default budgets are set via environment variables in the JupyterHub deployment:
| Variable | Description |
|---|---|
QUOTA_USER_CPU_S | Weekly CPU budget for regular users (in seconds) |
QUOTA_USER_GPU_S | Weekly GPU budget for regular users (in seconds) |
QUOTA_ADMIN_CPU_S | Weekly CPU budget for admins (in seconds) |
QUOTA_ADMIN_GPU_S | Weekly GPU budget for admins (in seconds) |
QUOTA_COLLAB_CPU_S | Weekly CPU budget for collaboration servers (in seconds) |
QUOTA_COLLAB_GPU_S | Weekly GPU budget for collaboration servers (in seconds) |
All default to 72 000 seconds (20 hours) if not set.
Server profiles
Each profile in the JupyterHub configuration belongs to a tier based on its name. Profiles whose name or slug contains “gpu” are assigned to the GPU tier; all others are CPU. Example:
profileList:
- display_name: "Minimal environment"
slug: mini-cpu
default: true
kubespawner_override:
node_selector:
sku: "cpu"
- display_name: "BME GPU Server (Dedicated GPU)"
slug: bme-gpu-dedicated
kubespawner_override:
extra_resource_limits:
nvidia.com/gpu: "1"
node_selector:
gpu-mode: "dedicated"
To add or modify profiles, edit config_jupyterhub.yaml.tmpl in the relevant deployment folder and trigger a re-deploy via the GitHub Actions workflow.