When it comes to data science and machine learning, notebooks (based on Jupyter) are often the main tool for research and exploration as they allow interactive work with data, in-line visualization, and co-coding. Data scientists may want to move away from notebooks running locally to a cloud service, especially when they need flexible and more robust infrastructure to host the notebooks and when they want to collaborate with others. Google Cloud offers two excellent options: Vertex AI Workbench and Colab Enterprise. Both are built on the popular Jupyter notebook platform, but each has its own strengths and ideal use cases. Last year, we showed how you can get away with almost free notebooks on GCP. In this post, we’ll compare these two solutions to help you decide which one suits your needs best.
📔 What They Have in Common: Jupyter Notebooks
Both Vertex AI Workbench and Colab Enterprise use the same core engine: Jupyter notebooks, a favorite among data scientists for their interactive capabilities. Jupyter notebooks allow you to write and execute code, visualize data, and document your findings all in one place. This shared foundation ensures that both tools offer a powerful and familiar interface for your data projects. However there are a few major differences.
Comparison Table
Feature | Colab Enterprise | Vertex AI Workbench |
|---|---|---|
Environment | Managed, collaborative | Customizable, developer-focused |
Infrastructure Management | Serverless, managed by Google | User-controlled, flexible |
Collaboration | Excellent, with IAM control | Good, with GitHub integration |
Compute Provisioning | Automatic | User-configurable |
Data Integration | Seamless with Google Cloud services | Seamless with Google Cloud services |
Code Completion | Inline | Inline |
Customization | Limited | Extensive |
GPU Support | ✓ | ✓ |
Conda Environments | ✘ | ✓ |
Custom Containers | ✘ | ✓ |
Automated Notebook Runs | ✘ | ✓ |
Idle Shutdown | Automatic | Configurable |
Persistent Storage | ✘ | ✓ |
Access to VM | ✘ | ✓ |
Original Jupyter UI | Modified | Retained |
Colab Enterprise: Ideal for Collaboration and Ease of Use
Let's start with Colab Enterprise. It is designed to make collaboration easy and free you from the hassles of managing infrastructure. Originally, it evolved from Google Colab, which was part of the Google Workspace (previously Gsuite) ecosystem, much like Google's version of Microsoft Office.
Key Features:
🔗 Share and Collaborate: Easily share notebooks with individuals, Google groups, or entire Google Workspace domains. Access control is handled through Google Cloud’s IAM.
🌐 Managed Compute: Colab Enterprise takes care of provisioning and managing compute resources. It starts runtimes when needed and shuts them down when not in use.
✅ Google Cloud Integration: Seamlessly work with Google Cloud services like Vertex AI and BigQuery from within your notebook.
✨ Inline Code Completion: Write code faster with suggestions that pop up as you type.
When working with Colab, you should consider the experience that Google intended to provide, which is somewhat similar to Google Docs/Slides. It's designed to be serverless and well-connected to your G Suite data (Drive, files, etc.). The concept of sharing and creating copies of the notebook is at its core. An ideal scenario is when you want to show your colleague some analysis you have done, allowing them to duplicate it and experiment with it themselves in a new environment.
However, this solution is less efficient when you want to run heavy workloads, as the runtime needs to be extended for long tasks, or when you want the data to persist on the disk of the machine once it's turned off (or released, in this case). When you want to control the environment and optimize it, you would typically prefer a more professional experience, which brings us to the next product: Workbench.
Vertex AI Workbench: Maximum Control and Customizability
Vertex AI Workbench is a full Google Cloud native product. Based on Deep Learning VMs the product offers customization options, making it better for those who need more control over the machine that runs the Jupyter environment.
Key Features:
👨🏻💻 Access to the VM: Unlike Colab Enterprise, you get full access to the virtual machine itself, allowing for in-depth configuration tailored to your specific needs. You can integrate more easily with your GCP environment based on IAM.
📦 Persistent Storage: Data isn't lost when the machine restarts, as the VM's disk is retained, ensuring your data remains intact.
☑ Controlling Instance Types: Choose from several types of instances, including N2 CPU or any GPU offering that GCP has.
🤏 Preinstalled Packages and GPU Support: All instances come with JupyterLab and a suite of deep learning packages like TensorFlow and PyTorch, with GPU support available.
</> GitHub Integration: Sync your notebooks with GitHub for version control and collaboration.
💾 Custom Environments and Containers: Add conda environments or create custom containers to tailor your setup to specific needs, so you don't need to install dependencies every time a team member wants to launch a new machine.
👾 Data Integration: Access Cloud Storage and BigQuery directly from JupyterLab by identifying either as the user working on the notebook or as a service account.
🛠️ Automated Notebook Runs and Idle Shutdowns: Schedule notebook runs and automatically shut down idle instances to manage costs effectively.
🖥️ Original Jupyter UI: Workbench retains more of the original Jupyter UI, providing a cleaner and more familiar interface for users accustomed to Jupyter notebooks.
Which One Should You Choose?
As you can see, for lightweight usage, both products can work fine. Colab is excellent when you want to collaborate, share code between team members, run ad-hoc analyses, and allow users in your organization to make copies of your notebook as they would collaborate around a Google Doc. Vertex AI Workbench is more of a cloud service, offering more control over the infrastructure, scalability, and the ability to maintain the environment. Therefore I'd say:
For Collaboration and Simplicity: If your priority is to collaborate easily with others and avoid managing infrastructure, Colab Enterprise is the way to go. It’s designed to make teamwork simple and setup effortless.
For Customization and Control: If you need detailed control over your environment and extensive customization options, Vertex AI Workbench is your best bet. It supports complex workflows and allows you to configure instances to meet specific requirements.
Summary
Both Colab Enterprise and Vertex AI Workbench are fair tools that integrate well with Google Cloud services. Your choice will depend on what your project needs: ease of collaboration and management, or deep customization and control. In conclusion, understanding the features and strengths of Colab Enterprise and Vertex AI Workbench can help you select the right tool for your data science and machine learning projects. Each solution offers unique benefits, so choose the one that best fits your workflow and goals.