Meta 发布自研推荐训练芯片 MTIA 300，集成网络通信
原标题：Meta Expands Its Custom Silicon Strategy From Compute Into Networking
AI 摘要
Meta 发布了其首款用于训练排序和推荐模型的自研加速器 MTIA 300，该芯片将网络和集合通信直接集成到芯片中，以解决推荐模型训练中通信开销大的问题。MTIA 300 包含两个网络 chiplet，提供 1.2 TB/s 的总 I/O 带宽，并配备 16 个专用消息引擎，使通信与计算并行，性能损耗低于 0.5%。Meta 还扩展了与博通的合作，计划未来两年推出四代新芯片，覆盖排序、推荐和生成式 AI 工作负载。
正文节选
Meta has detailed MTIA 300, its first in-house accelerator optimized for training ranking and recommendation models. Unlike large-language-model training, where raw floating-point throughput is often the dominant concern, Meta says recommendation models spend much more time communicating between accelerators. MTIA 300 addresses that by integrating networking and collective communication directly into the chip. The communication pressure comes largely from embedding tables, which Meta says can co