Notebooks & GPU¶
PAIS provides GPU-enabled Jupyter notebooks via Kubeflow Notebooks. Each research group gets isolated notebook servers with direct access to:
- NVIDIA L40S GPU (48 GB VRAM, time-sliced)
- VAST network-attached storage (shared across the group)
- Pre-installed ML frameworks (PyTorch, TensorFlow, HuggingFace)
- PAIS inference API (via environment variables)
Access¶
Notebooks are at https://kubeflow.pais.auckland.ac.nz. Log in with your University of Auckland account via Tuakiri.
Quick links¶
| Task | Where to look |
|---|---|
| Start a notebook | Launching a Notebook |
| Request GPU resources | GPU Access & Scheduling |
| Access datasets and save results | Working with Storage |
Default notebook images¶
| Image | Contents | Use case |
|---|---|---|
jupyter-pytorch-cuda |
PyTorch, CUDA, common ML libs | Model training, experimentation |
jupyter-tensorflow-cuda |
TensorFlow, Keras, CUDA | TF-based training |
jupyter-scipy |
SciPy stack, pandas, matplotlib | Data analysis, no GPU needed |
| Custom | Bring your own Dockerfile | Specialised environments |
Custom images can be built with buildah on the control plane. Contact the PAIS team.