How to Build a GPU‑Accelerated Jupyter Notebook Server with Docker for Google Colab

1. Introduction

The article explains how to create a personal GPU‑accelerated Jupyter Notebook server using Docker, enabling deep‑learning model training on a local machine or server without the resource limits of Google Colab or Kaggle.

2. Docker Overview

Docker runs applications in isolated containers that package all required dependencies, making them portable across hosts regardless of installed software. The client communicates with the Docker daemon, which manages images (blueprints) and containers (runnable instances). Images are pulled from the public registry when missing.

3. Enabling GPU Acceleration

Docker itself does not provide GPU access; the Nvidia Container Toolkit and CUDA drivers are required. The host must have Nvidia GPUs with the appropriate driver installed. The toolkit injects the CUDA runtime into containers, allowing GPU‑enabled applications to run.

4. Installation Guide

The following steps assume a Windows 11 host with WSL2 enabled.

Activate WSL2 and install Ubuntu.

PS> wsl --install -d Ubuntu

PS> wsl -l -v

PS> wsl --set-default-version 2

Install Docker Desktop and ensure it uses the WSL2 backend.

Enable GPU support by installing the Nvidia driver, CUDA toolkit, and Nvidia Container Toolkit inside the Ubuntu subsystem.

$ distribution=$( . /etc/os-release; echo $ID$VERSION_ID )
$ curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
$ curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list
$ sudo apt-get update
$ sudo apt-get install -y nvidia-docker2
$ sudo service docker stop
$ sudo service docker start

Test the GPU runtime with a benchmark container.

docker run --gpus all nvcr.io/nvidia/k8s/cuda-sample:nbody nbody -gpu -benchmark

Pull the official TensorFlow GPU‑Jupyter image.

docker pull tensorflow/tensorflow:latest-gpu-jupyter

5. Running the Notebook Server

Start a container from the TensorFlow image, exposing port 8888.

docker run --gpus all -p 8888:8888 -it --rm tensorflow/tensorflow:latest-gpu-jupyter bash

Inside the container, launch Jupyter with options that allow connections from Google Colab.

jupyter notebook --notebook-dir=/tf --ip 0.0.0.0 --no-browser --allow-root --NotebookApp.allow_origin='https://colab.research.google.com' --port=8888 --NotebookApp.port_retries=0

The command prints a secret token; the screenshot below highlights it.

6. Connecting Google Colab to the Local Runtime

In a Colab notebook, click the connection arrow next to the Connect button and choose “Connect to local runtime”. Paste the token (changing 127.0.0.1 to localhost) and click Connect.

After connection, the list of available devices shows the Nvidia GeForce RTX 2060 GPU.

7. Conclusion

Docker provides isolated containers for applications; adding the Nvidia CUDA Toolkit enables GPU access inside those containers. By pulling a TensorFlow GPU image that includes Jupyter, users can run a notebook server locally and connect Google Colab to it, gaining full control over the hardware and software environment for deep‑learning training.