What you'll learn

Retrieve real-time data on global energy mixes and carbon intensity from the ElectricityMaps API. Identify power grids that produce electricity from low-carbon sources, such as hydro, nuclear, wind, and solar power.
Run a machine learning training job using low-carbon electricity by re-directing training tasks to cloud server locations selected based on their average and real-time carbon intensity measurements.
Analyze the carbon footprint of sample Google Cloud usage data, including machine learning training, inference, storage, and other API activities.

Course recap

PRO

This course, built in partnership with Google Cloud, teaches developers how to reduce the carbon footprint of AI workloads by making informed decisions about where and when to run compute jobs. Cloud computing accounts for roughly 2.5-3.7% of global greenhouse gas emissions, and AI workloads are growing rapidly. By choosing data centers powered by low-carbon energy sources (wind, solar, hydro, nuclear) and timing jobs appropriately, developers can significantly reduce emissions without sacrificing performance.

Concept map

Key concepts

Carbon-free energy (CFE)Percentage of electricity from renewable or nuclear sources powering a data center
Carbon intensityAmount of CO2 emitted per unit of electricity generated, varying by region and time
Greenhouse Gas ProtocolCorporate standard for quantifying and reporting carbon emissions (Scope 1, 2, 3)
Regional selectionChoosing cloud data center regions based on carbon footprint, cost, and latency trade-offs
Inference vs training footprintInference energy costs can surpass training costs within weeks of deployment at scale

About this course

Learn how to perform model training and inference jobs with cleaner, low-carbon energy in the cloud!

Learn from Nikita Namjoshi, developer advocate at Google Cloud and Google Fellow on the Permafrost Discovery Gateway, and explore how to measure the environmental impact of your machine learning jobs, and also how to optimize their use of clean electricity.

Query real-time electricity grid data: Explore the world map, and based on latitude and longitude coordinates, get the power breakdown of a region (e.g. wind, hydro, coal etc.) and the carbon intensity (CO2 equivalent emissions per kWh of energy consumed).
Train a model with low-carbon energy: Select a region that has a low average carbon intensity to upload your training job and data. Optimize even further by selecting the lowest carbon intensity region using real-time grid data from ElectricityMaps.
Retrieve measurements of the carbon footprint for ongoing cloud jobs.
Use the Google Cloud Carbon Footprint tool, which provides a comprehensive measure of your carbon footprint by estimating greenhouse gas emissions from your usage of Google Cloud.

Throughout the course, you’ll work with ElectricityMaps, a free API for querying electricity grid information globally. You’ll also use Google Cloud to run a model training job in a cloud data center that is powered by low-carbon energy.

Get started, and learn how to make more carbon-aware decisions as a developer!

Who should join?

Familiarity with Python will help with the coding parts of the lesson, although the concepts covered can help anyone gain an understanding of the environmental impact of machine learning workflows.