Broadcom· @Broadcom · X·· 23 天前AI 评分33
AI 导读
传统的半导体缩放方法已不足以满足超大规模 AI 工作负载的需求。 在 TSMC OIP 生态论坛上，Greg Dix 将讨论 Broadcom 规模化定制 XPU 平台的方法，包括先进制程节点的采用、异构多芯片（multi-die）集成以及近存算架构协同设计。 欢迎参加他 9 月 23 日的主旨演讲：“构建 AI 未来：用明天的方案解决今天的 XPU 挑战。”
AI 生成摘要 · 以原文为准
正文
Traditional semiconductor scaling methods are no longer sufficient to meet the demands of hyperscale AI workloads.
At the TSMC OIP Ecosystem Forum, Greg Dix will discuss Broadcom’s approach to custom XPU platforms at scale, including advanced node adoption, heterogeneous multi-die integration, and near-memory architectural co-design.
Join his keynote on September 23: “Architecting the AI Future: Solving Today’s XPU Challenges with Tomorrow’s Solutions.”
来源：Broadcom · x.com