[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"aihot-art-89197":3},{"itemId":4,"vertical":5,"category":6,"source":7,"score":8,"title":9,"summary":10,"analysis":11,"url":12,"coverUrl":13,"direction":13,"marketSignal":13,"publishedAt":14},"89197","ai","技巧与观点","NVIDIA Technical Blog",60,"英伟达详解BlueField极致协同设计，支撑智能体AI工厂扩展","英伟达详解用BlueField芯片对智能体AI基础设施做协同设计,应对多步骤调用带来的数据搬运压力。","英伟达详解智能体AI工厂如何通过与BlueField DPU的极致协同设计来扩展,核心是智能体式工作负载让基础设施的调用模式变得远比传统推理复杂。\n\n· 智能体架构下,一次用户请求可能触发多次模型调用、工具调用、记忆查询、策略检查、存储访问和网络传输\n· 随着并发智能体数量增多,且要在多用户、多工具、多会话间携带上下文,基础设施必须更快地搬运、保护、检索和复用数据\n· BlueField作为数据处理单元被定位为承接这些高频、细粒度数据操作的关键组件,与计算侧协同设计以降低延迟\n· 文章聚焦架构理念,未给出具体性能数字或产品发布细节\n\n这透露出英伟达对智能体基础设施的下一步押注方向——不只是堆算力,而是把DPU这类数据搬运层也纳入协同设计,是判断未来AI工厂硬件路线的一个信号。","https:\u002F\u002Fdeveloper.nvidia.com\u002Fblog\u002Fscaling-agentic-ai-factories-through-extreme-co-design-with-nvidia-bluefield\u002F",null,"2026-07-17 00:00:00"]