Micron Technology· @MicronTech · X·· 15 天前AI 评分36
AI 导读
智能体 AI 的愿景很简单：花更少的时间管理任务，把更多时间专注于更高价值的工作。 正如美光核心数据中心业务部门高级副总裁兼总经理 Jeremy Werner 所解释的，智能体 AI 运行在两个计算栈上，而非一个。GPU 负责推理与规划，CPU 负责编排执行。KV cache 等技术让 AI 上下文保持可用，但不断增长的上下文规模和智能体工作负载，正使内存需求远超单纯推理所需。了解更多：https://bit.ly/4xN0UTo
AI 生成摘要 · 以原文为准
正文
The promise of agentic AI is simple: less time spent managing tasks and more time focused on higher-value work.
As Micron SVP & GM, Core Data Center Business Unit, Jeremy Werner explains, agentic AI runs on two compute stacks, not one. The GPU reasons and plans, while the CPU orchestrates actions. Technologies like KV cache keep AI context available, but growing context sizes and agentic workloads are driving memory needs far beyond reasoning alone. Learn more: https://bit.ly/4xN0UTo
来源：Micron Technology · x.com