NVIDIA Vera Rubin NVL72 achieves 30x higher throughput per megawatt via AgentX
August 29, 2026
NVIDIA's Vera Rubin NVL72 architecture demonstrates up to 30x better throughput per megawatt compared to GB300 NVL72 when running the SemiAnalysis AgentX workload. This benchmark focuses on energy-efficient throughput specifically for agentic workflows rather than traditional training metrics.
HOW THIS AFFECTS YOU
●
builderYou can optimize infrastructure selection based on agent-specific energy efficiency rather than raw FLOPS.
●
investorThis signal highlights the shift toward power-efficiency as a primary competitive metric in AI data center hardware.