NVIDIA's GB300 NVL72 Dominates First Agentic AI Benchmark, Running 20x More Agents Per Megawatt Than Previous Generation
Summary
NVIDIA's GB300 NVL72 dominates the industry's first agentic AI benchmark, AgentPerf, delivering a staggering 20x more AI agents per megawatt than its predecessor, the HGX H200, as top inference providers race to deploy the powerful Blackwell architecture for real-world production applications.
Key Points
- NVIDIA's GB300 NVL72 tops the industry's first agentic AI benchmark, AgentPerf, running up to 20x more agents per megawatt than the NVIDIA HGX H200 system on DeepSeek V4 Pro workloads.
- AgentPerf, developed by Artificial Analysis, measures real-world agentic AI performance by simulating coding agent tasks across 12+ programming languages, tracking how many concurrent agentic tasks a platform can handle while meeting responsiveness thresholds.
- Leading inference providers including Baseten, DeepInfra, and Together AI are already deploying agentic workloads on NVIDIA Blackwell, powering production applications like Cursor's AI coding platform and Pam.ai's automotive AI workforce solution.