A Practical Guide to Running AI/ML Experiments on GCP

A hands-on tutorial for beginners who want to use Google Cloud Platform (GCP) for large-scale machine learning experiments or AI workloads.

🎯 What You’ll Learn¶

How to use the GCP console and the gcloud CLI
How to provision and manage virtual machines and GPU resources on GCP
How to run large-scale ML experiments (like fine-tuning large language models) on GCP VMs
When to use virtual machines vs. managed services like Vertex AI
How to predict and manage GPU memory
How to avoid expensive cloud mistakes with cost monitoring and budget alerts

📋 Prerequisites¶

Required:
- Python programming
- Comfort with terminal/command line
- Basic knowledge of machine learning/deep learning
Not required:
- GCP experience (we start from scratch!)
- Deep learning expertise
- Cloud computing background
You will also need:
- A Google Cloud account (there is a free tier available with $300 credits)
- A text editor (VS Code, Sublime Text, Cursor, etc.)
- Terminal access

🕥 Duration¶

4-6 hours (can be split into multiple sessions)

Core tutorial: 3-4 hours
Optional appendices: 25-30 minutes additional

📚 Tutorial Structure¶

Core Path¶

Part	Title	Time	Learning Objective
1	GCP Foundations	30 mins	Understand GCP and interact with it via the console and `gcloud` CLI
2	Running Experiments on GCP VMs	45 mins	Learn how to run AI/ML experiments on GCP virtual machines with GPU support
3	Vertex AI	65 mins	Understand how to use the managed Vertex AI service for training jobs and notebooks
4	Cost Management	30 mins	Understand budget alerts and best practices for cost management

Appendix (Optional)¶

Appendix	Title	Time	Focus
A	GPU Memory Maths	45 mins	Understand how to estimate GPU memory requirements, avoid OOM errors, and choose the right GPU for your needs

💡 What Makes This Tutorial Different¶

This tutorial teaches you what you need to run AI/ML experiments on GCP, emphasizing:

Systems thinking for ML infrastructure
Cost-conscious cloud practices
How to structure reproducible AI/ML experiments for real-world projects
Hands-on practice with every learned concept

The content is also structured to be used on-demand so you do not have to read the entire tutorial.

🎓 Who This is For¶

This tutorial was particularly designed for AIMS AI for Science students who do not have a background in cloud computing or software engineering, helping them use GCP more effectively for their research projects.

However, it should be generally helpful for:

Beginner AI/ML students and researchers
Anyone curious about using GCP for running AI/ML experiments

🧠 How to Use This Tutorial¶

Option 1: Linear (Recommended for beginners)¶

Start from Part 1 and work through sequentially.

Option 2: Jump to What You Need¶

Need a quick-start guide to using GCP? → Part 1
Want to run ML experiments on a GPU VM? → Part 2
Want to learn how to use managed services like Vertex AI? → Part 3
Need to understand how to monitor costs to manage your GCP credits properly? → Part 4
Need to know how to choose a GPU for your experiment? → Appendix A

🫱🏿‍🫲🏾 Contributing¶

This tutorial is fully open-source! We welcome:

🐛 Bug reports
💡 Suggestions for improvements
📝 Additional sections, examples, or exercises
🌎 Translations

See CONTRIBUTING.md for details.

📄 License¶

This tutorial is released under the CC-BY-4.0 License. See LICENSE for details.

Let’s get started! → Part 1: GCP Foundations