Breaking-down IT hardware component emissions for sustainable decision-making at scale

Описание: Lisa Rivalin (Meta - Production System Engineer)

Emre Tepedelenlioglu (Meta - Production Systems Engineer)


The embodied emissions of Meta's IT hardware fleet are influenced by various factors throughout the hardware's lifecycle- from design and procurement to decisions such as determining the useful life of systems as a whole and their components. We have developed a scalable methodology using GenAI to assess the carbon impact of components at various levels of abstraction. This enables us to break down emissions into views along two axes: domain (training- network- compute...) and component (CPU- GPU- Flash...). This approach allows us to accurately identify emissions associated with each part and enables infrastructure teams to understand the emission profiles of IT hardware from a single part to the entire fleet. Our presentation includes a description of the methodology and a fleet view- which provides an understanding of the evolution of the emissions landscape. We will also zoom into emissions at the component level- identifying areas where actions can be taken to reduce emissions.