英伟达调整Vera Rubin量产规格
AI 总结
英伟达称,修订后的Vera Rubin规格代表系统通过认证并进入量产后可维持的实际指标,而非开发阶段公布的初步架构目标。调整后,每颗Rubin GPU的HBM4带宽从22TB/s降至19.2TB/s,NVLink 6带宽从3.6TB/s降至3TB/s,并适用于该平台全部配置。英伟达还表示Vera Rubin NVL72已经发货,已发布的第三方智能体测试采用量产规格,但现有报道未独立验证这些厂商说法。
深度见解
此次更新为Vera Rubin这一重要AI基础设施平台提供了更准确的量产指标,也说明了英伟达如何从系统层面评估其智能体工作负载表现。
NVIDIA’s Official Statement on Vera Rubin Spec Change: Updated Specs “Reflect What The System Sustains At Quality, Volume Production”, And All 3rd Party Benchmarks Have Been Measured Using Production Specs
Hassan Mujtaba•Oct 29, 2026 at 03:10am EDT
Add to Preferred Sources
NVIDIA has clarified why it updated specifications for Vera Rubin, and says that they better reflect what the systems sustain at volume production.
部分来源内容未展示,可前往原文继续阅读。
#NVIDIA#Vera Rubin#Rubin GPU#Vera Rubin NVL72#HGX Rubin NVL8#GB300 NVL72#HBM4#NVLink 6#AgentX#SemiAnalysis#AgentPerf#Artificial Analysis