Skip to main content

2 - Access Jupyter Notebook via Browser

Why this guide's resources? The Jupyter server runs on the GPU VM because that is where the GPU, driver and packages are; this guide only makes it reachable. One firewall rule opens exactly port 8888 so your browser can connect, and the token in the URL is the only secret between you and the machine. See 0 - The Cloud Resources in This Tutorial for why each piece is there.

Goal: Configure firewall rules and access the Jupyter notebook running on your GPU instance through a web browser. This provides the simplest access method and works in any web browser.

Advantages of browser-based Jupyter:

  • No local software installation required
  • Works on any device with a web browser
  • Quick setup - just open firewall and access via URL
  • Perfect for Chromebooks and mobile devices

Prerequisites​

  • Completed tutorial 1 (GPU instance created, Julia packages installed, Jupyter kernels registered)
  • Jupyter not yet running on the server

Step 1: Launch Jupyter notebook securely​

First, make sure the notebook server is installed on the instance (tutorial 1 installs it, but if you skipped that part of the setup, install it now):

sudo apt-get install -y jupyter-nbconvert jupyter-notebook

jupyter-notebook (6.4.8 on Ubuntu 22.04) provides the notebook server; jupyter-nbconvert alone is NOT sufficient because IJulia's notebook(...) launch requires the notebook subcommand.

Kernel name for nbconvert: if you execute the notebook headless with jupyter nbconvert --execute, pass --ExecutePreprocessor.kernel_name=julia-1.11 so it matches the installed kernel (jupyter kernelspec list). If the notebook metadata and the installed kernel name differ, fix the metadata or pass the name that is actually installed.

Then ask Gemini to start Jupyter with authentication enabled:

SSH into the server and launch Jupyter notebook using IJulia so it is accessible over the network and runs in the background.

Julia command:

julia -e 'using IJulia; notebook(`--ip=0.0.0.0 --no-browser`, dir=pwd(), detached=true)'

Then ask Gemini to retrieve the Jupyter URL with token:

Show me the Jupyter server URL including the token from the running Jupyter process logs.

Security note: Jupyter auto-generates a secure token. The full URL including the token will be printed in the logs — only share it with trusted users.

Open firewall port for Jupyter​

Step 2: Allow Jupyter port access​

Ask Gemini to create a firewall rule to allow access to Jupyter:

Create a firewall rule to allow access to port 8888 for Jupyter notebook. The rule should allow TCP traffic from any IP address.
Then show the full connection URL for Jupyter including the external IP and the token.

Expected firewall rule creation:

gcloud compute firewall-rules create allow-jupyter \
  --allow tcp:8888 \
  --source-ranges 0.0.0.0/0 \
  --description "Allow Jupyter notebook access"

Gemini will also provide the connection URL with a token, typically looking like:

http://xxx.xxx.xxx.xxx:8888/?token=abcdef1234567890...

Note: If the firewall creation says the rule allow-jupyter already exists, that is fine: the project already has the rule, so just continue (do not create a duplicate).

Step 3: Open in the browser​

Paste the URL from the previous step into your web browser to access the Jupyter notebook interface.

Important: Don't share the token URL with others. The token provides secure access to your Jupyter notebook and GPU instance.

Step 4: Verify GPU access​

In Jupyter, create a new Julia notebook and run:

using CUDA
CUDA.versioninfo()

Expected output should show:

CUDA toolchain: 
- runtime 13.2, artifact installation
- driver 580.178.4 for 13.3
- compiler 13.3, artifact installation

CUDA libraries: 
- CUBLAS: 13.4.0
- CURAND: 10.4.2
- CUFFT: 12.2.0
- CUSOLVER: 12.2.0
- CUSPARSE: 12.7.10
- CUPTI: 2026.1.1 (API 13.2.1)
- NVML: 13.0.0+580.178.4

Julia packages: 
- CUDA: 5.11.3
- GPUArrays: 11.5.16
- GPUCompiler: 1.17.1
- KernelAbstractions: 0.9.45
- CUDA_Driver_jll: 13.3.5+0
- CUDA_Compiler_jll: 0.4.4+1
- CUDA_Runtime_jll: 0.21.0+1

Toolchain:
- Julia: 1.11.9
- LLVM: 16.0.6

1 device:
  0: Tesla T4 (sm_75, 14.559 GiB / 15.000 GiB available)

(These are the versions observed on Ubuntu 22.04 LTS as of October 2026; exact patch versions of CUDA.jl and friends may differ slightly on a fresh install, which is fine.)

Upload the tutorial notebook​

If you followed tutorial 1, the repo is already cloned on the server (in ~/cloudbank_showcase/gpu_computing_oceanography/) and you can skip this section. If you did not clone it, upload the notebook files from your local machine:

  1. In Jupyter, click Upload in the top right corner
  2. Select the ocean_gpu_tutorial.ipynb, ocean_utils.jl, and Project.toml files from your local machine
  3. Click Upload to confirm
  4. Open the tutorial notebook to begin the ocean modeling tutorial

Security considerations​

Important: The firewall rule allows access from any IP address. When you're done working:

  1. Delete the firewall rule: 
    gcloud compute firewall-rules delete allow-jupyter
  2. Stop the instance if not needed: 
    gcloud compute instances stop julia-ocean-gpu --zone=us-west1-b

Troubleshooting​

  • Connection refused: Ensure Jupyter is running and the firewall rule is in place
  • Token not found: Ask Gemini to show the latest Jupyter logs to find the correct token
  • Slow performance: This is normal during first GPU compilation; subsequent runs will be much faster

Alternative access method​

If you prefer using VS Code instead of browser-based Jupyter for a full IDE experience, see tutorial 3_access_jupyter_vscode.md.