Posted 4/10/2024, 3:05:08 PM
Meta Unveils Next-Gen AI Chip, Doubling Performance for Recommendation Models
- Unveiled next generation AI inference accelerator (MTIA v2) that more than doubles compute and memory bandwidth of previous version
- MTIA v2 designed specifically for ranking and recommendation models that provide recommendations to users
- MTIA v2 now deployed in data centers and serving models in production, allowing more investment in intensive AI workloads
- MTIA v2 achieving greater efficiency than commercially available GPUs since Meta controls the whole stack
- This is part of a long-term roadmap to build the most efficient infrastructure possible for Meta's unique AI workloads