A new technical paper titled "Life-Cycle Emissions of AI Hardware: A Cradle-To-Grave Approach and Generational Trends" was published by researchers at Google.
The paper presents a comprehensive life-cycle assessment (LCA) of greenhouse gas emissions for AI hardware accelerators, including manufacturing emissions.
The study introduces a new metric called compute carbon intensity (CCI) to evaluate AI hardware sustainability and estimates the carbon footprint of training and inference.
The research highlights the improvements in CCI from TPU v4i to TPU v6e, and emphasizes the importance of software advancements in leveraging and amplifying these gains.