NVIDIA Vera Rubin 和 Blackwell 树立代理式 AI 每瓦性能新标准
原标题：NVIDIA Vera Rubin and Blackwell Set a New Standard for Agentic AI Performance per Watt
AI 摘要
NVIDIA 技术博客介绍了 SemiAnalysis 的 AgentX 基准测试，用于评估 AI 基础设施在代理式编码推理中的性能。该基准通过重放真实编码代理会话，测量每兆瓦吞吐量等指标。NVIDIA 公布的预览结果显示，Vera Rubin NVL72 在 AgentX 工作负载下每兆瓦吞吐量比 GB300 NVL72 高出 30 倍，而 GB300 NVL72 比 H200 NVL8 高出 15 倍，展示了新一代硬件在代理式 AI 推理中的能效优势。
正文节选
AI agents have expanded inference from single-turn interactions into multi-step workflows that reason, invoke tools, coordinate subagents, and carry growing context from one turn to the next. The scale of this shift is now visible in raw consumption: across 100 trillion tokens of real-world usage, OpenRouter’s State of AI report found that average prompt tokens per request grew roughly fourfold, and single agentic requests consume 15 times the tokens of ordinary chat. Properly characterizing har