NVIDIA Blackwell Leads on First Agentic AI Infrastructure Benchmark
NVIDIA Blog · Shruti Koparkar · 2026-06-12
NVIDIA's Blackwell Ultra NVL72 platform leads AgentPerf, Artificial Analysis's first agentic AI infrastructure benchmark, delivering 20x more agents per megawatt than NVIDIA's prior Hopper generation when running DeepSeek V4 Pro across chained multi-step agentic workloads.
Appears in
Extraction
Topics: agentic-ai-infrastructureai-benchmarksnvidia-hardwareinference-performance
Claims
- AgentPerf is the first benchmark designed specifically for agentic AI workloads, measuring concurrent agent throughput and efficiency rather than single LLM call speed.
- NVIDIA GB300 NVL72 runs up to 20x more agents per megawatt than the NVIDIA HGX H200 on DeepSeek V4 Pro workloads.
- Agentic workloads involve dozens to hundreds of chained LLM and tool calls, making their complexity multiplicative rather than additive compared to single completions, which existing benchmarks fail to capture.
- Production agentic applications including Cursor and Pam.ai are already running on Blackwell hardware through providers including Together AI and DeepInfra.
Key quotes
An agent functions more like a relay: It breaks a goal into many steps and keeps going until the task is done.
The complexity isn't additive; it's multiplicative.
NVIDIA GB300 NVL72 delivers the highest performance in the benchmark, running up to 20x more agents per megawatt than the NVIDIA HGX H200 system.