Nvidia's DSX Platform Boosts AI Data Centre Output with Power Management Software

Nvidia has released early deployment results for its DSX software and infrastructure platform designed for AI data centres. The company claims that electricity, rather than chips alone, is becoming the main constraint on AI data centre expansion.

The DSX software includes two approaches to power management: shifting computing workloads when the grid is under strain, and reallocating power across servers to raise output without increasing a site's total energy budget. In one example, Silicon Valley Power sent a signal to an AI facility in Santa Clara to cut electricity use during high demand periods.

The Emerald AI Conductor software successfully reduced the site's power draw from four megawatts to three while keeping higher-priority inference jobs running. The company has sent more than 200 demand signals to that facility, with each one triggering a successful automated response.

Nvidia also highlighted its DSX MaxLPS software, which monitors GPU and rack-level power use and shifts available headroom across systems according to workload demands. In tests on a five-rack, 19-node cluster, Lambda reported a 24% increase in cluster-wide token throughput and a 23% improvement in performance per watt.

Nvidia argues that optimising isolated parts of a facility is no longer enough as AI systems scale, and that a whole-facility approach is needed to increase useful work from each megawatt consumed. The company's DSX platform includes software for lifecycle management, simulation tools, and reference designs covering operations, facility planning, and power delivery.