NVIDIA DSX Increases AI Factory Token Throughput by 24% Within Fixed Power Budget

NVIDIA DSX is a technology suite that optimizes throughput per megawatt for AI factories. Cloud provider Lambda verified DSX MaxLPS on NVIDIA HGX B200 GPU servers, finding that operating 19 nodes within the power budget of 16 nodes increased overall cluster token throughput by 24% (from approximately 4 million tokens per second to 5 million tokens per second), while improving performance per watt by 23%. CEO Jensen Huang emphasized the importance of full-system optimization, including networking, cooling, and power distribution, stating, "A 1GW factory cannot become a 2GW factory." Additionally, Emerald AI's Conductor platform demonstrates how AI factories can serve as flexible grid resources by deprioritizing low-priority tasks within one minute in response to grid signals, reducing power usage from 4MW to 3MW. NVIDIA predicts that applying DSX MaxLPS to the next-generation Vera Rubin NVL72 factory could increase GPU capacity by up to 40% within the same power budget.