NVIDIA与Artificial Analysis推出AgentPerf：首个面向智能体AI基础设施的基准测试
英文摘要
NVIDIA and Artificial Analysis have released AgentPerf, the first benchmark designed specifically for agentic AI infrastructure. Unlike traditional benchmarks, AgentPerf measures performance when an AI agent chains dozens to hundreds of model calls, uses tools, gathers context, and iterates until task completion. The initial results highlight that NVIDIA Blackwell delivers 20 times more agents per megawatt compared to the previous NVIDIA Hopper architecture.
中文摘要
NVIDIA与Artificial Analysis发布了AgentPerf，这是首个专为智能体AI基础设施设计的基准测试。与传统的基准测试不同，AgentPerf衡量AI智能体在链接数十到数百次模型调用、使用工具、收集上下文并迭代直至任务完成时的性能。初步结果显示，NVIDIA Blackwell每兆瓦特处理的智能体数量比前代NVIDIA Hopper高出20倍。
关键要点
AgentPerf is the first benchmark tailored for agentic AI workloads, where AI agents make multiple model calls, use tools, and iterate.
AgentPerf是首个针对智能体AI工作负载量身定制的基准测试，AI智能体在此过程中会进行多次模型调用、使用工具并迭代。
The benchmark was developed by Artificial Analysis and gives developers, enterprises, and infrastructure providers a standard way to compare accelerated computing systems for agentic AI.
该基准测试由Artificial Analysis开发，为开发者、企业和基础设施提供商提供了一种比较智能体AI加速计算系统的标准方法。
Initial results show NVIDIA Blackwell achieves 20x more agents per megawatt than NVIDIA Hopper.
初步结果显示，NVIDIA Blackwell每兆瓦特处理的智能体数量是NVIDIA Hopper的20倍。
The announcement positions Blackwell as a key platform for energy-efficient agentic AI deployments.
该公告表明Blackwell是能效高效部署智能体AI的关键平台。