Free, hands-on labs for Kubernetes for AI infrastructure, from Cloud Brewery Academy and the Kubernetes Labs and Concepts newsletter. One concept per lab: break it, fix it, then show the evidence.
Independent course. Not affiliated with or endorsed by NVIDIA.
| Mode | Runs on | Needs |
|---|---|---|
| Laptop (default) | your machine | Docker (Docker Desktop or Colima), kind, kubectl; Helm from Lab 2 |
| Cloud GPU (optional parts) | one GPU instance in AWS that terminates itself after 55 minutes | an AWS account, GPU quota and a budget alarm; marked clearly in the lab that needs it |
Most labs run on a laptop. When a concept only shows up on real GPUs, that lab says so at the top.
| Lab | What you'll do | Mode |
|---|---|---|
| Lab 1: Schedule a GPU workload without a GPU | Make a Pending GPU pod schedule, keep other pods off GPU nodes, cap a team's GPUs, rehearse a GPU fleet with KWOK | Laptop |
| Lab 2: Share one GPU between pods with DRA | Share one (mock) GPU through a ResourceClaim, take slices of it, then see time-slicing vs MPS on a real GPU | Laptop; optional AWS GPU |
Read lab01-gpu-scheduling/README.md and follow the steps, or run
lab01-gpu-scheduling/scripts/run-lab.sh to do them all and save the evidence.
Code is licensed under Apache-2.0; lab guides and other written and visual material under CC BY 4.0. Third-party material keeps its own terms. See LICENSING.md for what applies where.