Skip to main content
Run JupyterLab on a dedicated cloud GPU. Train models, explore datasets, and prototype backed by an RTX 4090, A100, or H100 — without installing CUDA locally.

​
1. Deploy a devbox

runcrate instances create --name jupyter --gpu RTX4090 --template ubuntu-devbox
runcrate instances status jupyter

​
2. Install and start JupyterLab

runcrate ssh jupyter -- "pip install jupyterlab ipywidgets"

runcrate ssh jupyter -- "nohup jupyter lab \
  --ip=0.0.0.0 --port=8888 --no-browser --allow-root \
  --NotebookApp.token='' --NotebookApp.password='' \
  > /root/jupyter.log 2>&1 &"

​
3. Connect via browser

runcrate instances info jupyter
Open http://<INSTANCE_IP>:8888 in your browser.

​
4. Verify GPU access

In a notebook cell: 
import torch
print(f"CUDA: {torch.cuda.is_available()}")
print(f"GPU: {torch.cuda.get_device_name(0)}")
print(f"VRAM: {torch.cuda.get_device_properties(0).total_mem / 1e9:.1f} GB")

​
5. Persist notebooks with a volume

runcrate volumes create --name notebooks --size 50
runcrate instances create --name jupyter --gpu RTX4090 --template ubuntu-devbox --storage notebooks
Notebooks saved to /workspace/ persist across deploys.

​
Upload data

runcrate cp ./dataset.csv jupyter:/workspace/
runcrate cp ./my_notebook.ipynb jupyter:/workspace/

​
Install additional packages

runcrate ssh jupyter -- "pip install pandas scikit-learn matplotlib transformers"

​
Tips

  • Use the ubuntu-devbox template — it includes CUDA, cuDNN, and Python.
  • Save notebooks to /workspace/ when using a volume for persistence.
  • For long training runs, use nohup in a terminal tab so the job survives browser disconnects.
  • For password-protected access, run jupyter lab password and restart without the token flags.

​
Cleanup

runcrate instances delete jupyter