Nvidia's DSX MaxLPS Touts Dynamic Power Management for Data Centers

Nvidia's DSX MaxLPS site power management approach is designed to get more compute out of fixed data center power budgets. The company claims that its Vera Rubin NVL72 systems can deliver up to 2 zettaFLOPS for NVFP4 inference and up to 1.4 zettaFLOPS for NVFP4 training within a 100MW facility power budget.

Nvidia's approach uses dynamic power management resources, which are continuously aware of power usage at the chip level, rack level, and groups-of-racks level. This allows unused power to be redistributed to systems where it's most needed in the moment, maximizing performance per watt from the facility over time.

The company also notes that designing a data center with MaxLPS from the start grants an operator greater flexibility over the life of the installation. For example, if a site starts as a training-focused facility outfitted with cutting-edge hardware, each installed system is likely to need a greater share of the available site power for that more intense workload.

Nvidia's DSX MaxLPS approach also includes the use of higher liquid coolant temperatures for the exclusively liquid-cooled Rubin NVL72 racks. This allows for more power to be put to productive use, improving a site's power usage effectiveness (PUE) figure.