Nvidia unveils DSX MaxLPS power scheduling system, boosting token throughput per megawatt by up to 40%

At the AI Infrastructure Summit, Nvidia showcased its DSX MaxLPS power scheduling system for AI factories, boosting token throughput per megawatt by up to 40%. This metric measures how many tokens an AI system can generate or process per second for every 1 MW of power consumed, making it better suited for evaluating output under power-constrained conditions.

DSX MaxLPS has been integrated by Emerald AI into the Conductor platform for data center load orchestration under grid conditions. Its DSX Flex feature dynamically adjusts power demand based on grid status, and the platform has received 200 grid status signals without requiring user intervention.

Tests by cloud providers like Lambda showed that under a fixed power budget, a 19-node cluster's throughput rose from about 4 million tokens per second to 5 million, a 24% increase, with 23% better performance per watt. Nvidia also revealed that Vera Rubin and Groq 3 LPX racks can support up to 40% more GPUs within the same site power range, boosting token throughput by up to 35%.