Uploaded June 2023 | Updated September 2026, 6 hours ago

---

GPUs were not always optimal for running our specific recommendation workloads at the efficiency levels demanded by our scale, so we designed MTIA for Meta’s internal workloads ➡️ bit.ly/3IUyYXd

A closer look from Mahesh Maddury, Roman Levenstein & Linda Cheng; part of the team behind this custom silicon work at Meta.

--

Subscribe: youtube.com/facebookai?sub_confirmation=1

Learn more about our work: ai.facebook.com

Follow us on Twitter: twitter.com/AIatMeta
Follow us on Facebook: facebook.com/AIatMeta
Connect with us on LinkedIn: linkedin.com/showcase/AIatMeta

Working on AI at Meta, we're bringing the world together by advancing AI, powering meaningful and safe experiences, and conducting open research.