NVIDIA unveils dsx maxlps technology for enhanced AI data center efficiency

NVIDIA showcased its DSX MaxLPS power scheduling system at the AI Infrastructure Summit, claiming it can boost token throughput per megawatt by up to 40% in AI factories. The technology, integrated into Emerald AI’s Conductor platform, enables dynamic load management based on grid conditions. Lambda, a GPU cloud provider, reported a 24% increase in token throughput under fixed power budgets, with a 23% improvement in performance per watt. NVIDIA also noted that Vera Rubin and Groq 3 LPX racks could see up to a 40% increase in GPU count and a 35% rise in token throughput without additional power lines. Multiple AI firms tested early versions of Vera CPU, reporting performance gains of up to 1.9x, throughput increases of 73%, and query throughput improvements of 3.3x. Redpanda observed a 5.5x reduction in latency and 73% higher throughput, while Starburst and Kinetica saw query performance improvements of 3x and 2.7x respectively.