{
  "id": 3529783,
  "title": "Hot Chips 2026: Nvidia touts benefits of its DSX MaxLPS site power management approach — tech allows for more compute from fixed data center power budgets",
  "url": "https://urgent.news/2026/08/26/hot-chips-2026-nvidia-touts-benefits-of-its-dsx-maxlps-site-power",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-08-26T14:42:10.000Z",
  "source": {
    "name": "Tom's Hardware",
    "slug": "tom-s-hardware",
    "url": "https://www.tomshardware.com/tech-industry/data-centers/hot-chips-2026-nvidia-touts-benefits-of-its-dsx-maxlps-site-power-management-approach-tech-allows-for-more-compute-from-fixed-data-center-power-budgets"
  },
  "original_language": "en",
  "account": "In a presentation at Hot Chips 2026, Nvidia showcased its DSX MaxLPS site power management approach, highlighting the benefits of this technology in maximizing compute power within fixed data center power budgets. The ultimate constraint on performance in rack-scale AI systems is the amount of power available in data centers. Nvidia emphasized this constraint during its presentation of the Rubin GPU and stressed the importance of efficient power management for optimal productivity.\n\nThe company's DSX toolkit, which includes MaxLPS, allows operators to provision installations of 40,000 GPUs (or approximately 40 Rubin DGX SuperPODs) within a 100MW power budget. By utilizing all of Vera Rubin’s power management technologies in tandem with MaxLPS, Nvidia expects these hardware installations to deliver up to 2 ZFLOPS (NVFP4 inference) and up to 1.4 ZFLOPS (NVFP4 training).\n\nNvidia's back-of-the-napkin calculations suggest that the provided performance figures are estimates rather than measured results. However, the overall point remains: getting the most compute out of limited power budgets for future AI data centers will demand more refined planning, monitoring, and facility management than simply applying a coarse estimate of peak power draw for every electrical component.\n\nHistorically, data center operators have made assumptions about power usage based on worst-case power peaks per rack, leading to inflated power budgets that often result in underutilized resources. The DSX MaxLPS approach addresses this limitation by employing an intelligent, dynamic scheme that continuously monitors power usage at the chip, rack, and group-of-racks levels. This allows Nvidia's Dynamic Power Software control loop to redistribute unused power to systems where it's most needed, maximizing the number of systems that can be installed and improving performance per watt.\n\nIn a measured example using current GB300 racks and a 540kW power budget, static provisioning could accommodate four 135kW systems, while actual utilization might leave up to 170kW of the budget unused. Modern rack-scale systems like Nvidia’s NVL72s could safely install an entire additional rack within the same power budget, demonstrating the potential for higher performance per watt, increased rack installations, and greater revenue opportunities for data center operators.",
  "summary": "During Nvidia's Hot Chips presentation on the Rubin GPU, the company emphasized power as a hard limit on data center capacity and touted the amount of compute that Vera Rubin NVL72 systems can deliver within an example fixed facility power budget of 100MW.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "The Economic Times",
        "title": "AMI orders 9,000 Nvidia Vera Rubin chips",
        "url": "https://urgent.news/2026/08/26/ami-orders-9-000-nvidia-vera-rubin-chips",
        "published": "2026-08-26T00:30:00.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}