rapidsai/jupyterlab-nvdashboard

JupyterLab NVdashboard – at a glance

What it is – A JupyterLab extension that adds a live dashboard showing NVIDIA‑GPU metrics (utilisation, memory, PCIe/NVLink throughput) right inside your notebooks. It also ships a toolbar button that can switch on‑the‑fly GPU‑accelerated versions of pandas, scikit‑learn, UMAP or HDBSCAN (via RAPIDS cuDF and cuML) without changing your code.

Why it matters – When you’re developing data‑science notebooks on a GPU‑enabled machine, you normally have to leave the notebook to run nvidia‑smi or other monitoring tools. This extension puts those charts in the same UI, lets you brush time‑series, see synced tooltips across charts, and works with both light and dark JupyterLab themes. The “GPU Accel” button makes it easy to try RAPIDS‑accelerated libraries and instantly see the impact on the dashboard.

Key features

• Real‑time charts for compute utilisation, memory use, PCIe and NVLink throughput.
• Brushable time‑series and synchronized tooltips for easy comparison.
• Full theme‑compatibility with JupyterLab’s light/dark modes.
• One‑click activation of cuDF‑pandas and cuML accelerators (shown only when the corresponding RAPIDS packages are installed).
• Works only with JupyterLab v4+.

Getting it – Install via Conda (rapidsai or conda‑forge channels) or PyPI. Nightly builds are available for early testing.

Typical workflow

1. Open a notebook in JupyterLab v4.
2. Open the GPU Accel menu in the notebook toolbar and enable cuDF‑pandas or cuML.
3. Run your pandas / scikit‑learn code – the dashboard lights up with GPU activity and you can compare runtimes.
4. Use the brush tool on any chart to zoom into a time window; tooltips stay aligned across all charts.

Who should use it – Data scientists, ML engineers, or researchers who develop notebooks on NVIDIA GPUs and want instant visual feedback on hardware utilisation, or who want a painless way to try RAPIDS‑accelerated libraries.

---

Quick start (conda)

conda install -c rapidsai -c conda-forge jupyterlab-nvdashboard   # stable
# or nightly
conda install -c rapidsai-nightly -c conda-forge jupyterlab-nvdashboard

Quick start (pip)

pip install jupyterlab-nvdashboard               # stable
# or nightly
pip install --extra-index-url https://pypi.anaconda.org/rapidsai-wheels-nightly/simple 'jupyterlab-nvdashboard>=0.14.0a0'

---

Troubleshooting – Verify the server extension (jupyter server extension list) and the frontend labextension (jupyter labextension list) are both enabled.

---

The project is a genuine AI‑related tool: it helps developers monitor and accelerate GPU‑based data‑science workloads inside JupyterLab.