In a groundbreaking evaluation for agentic AI, the first-ever benchmark designed specifically for this emerging workload has been released. Called AgentPerf, it was developed by Artificial Analysis to help developers, enterprises, and infrastructure providers compare systems for agentic AI tasks. The initial results show that NVIDIA's Blackwell Ultra NVL72 platform delivers leading performance, running 20 times more agents per megawatt than its predecessor, the NVIDIA Hopper.
Agentic AI, which involves AI systems taking actions and making decisions autonomously, is fundamentally different from conversational AI. The benchmark methodology was built from the ground up to reflect how agentic AI operates in real-world production environments.
For a deeper dive into AgentPerf's methodology and NVIDIA's full-stack optimizations for agentic AI, refer to the technical blog from NVIDIA: Read more.