Uploaded June 2023 | Updated September 2026, 4 hours ago

---

More details on MTIA, Meta Training and Inference Accelerator ➡️ bit.ly/3IUyYXd

A closer look at the architecture of Meta's first-generation AI inference accelerator. MTIA v1 is a custom-designed ASIC built with next-generation recommendation model requirements in mind.

The accelerator is built on TSMC's 7nm process at a 25W TDP and provides 102.4 TOPS at INT8 precision and 51.2 TFLOPS at FP16 precision.

We're sharing more details on this work at #ISCA2023.

--

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.