{
  "schemaVersion": 1,
  "id": "atlas-lesson-rack-dynamic-power",
  "title": "Dynamic load, reserve and power control",
  "category": "learning",
  "format": "Atlas lesson",
  "description": "Accounts for changes in electrical demand over timescales from chip activity to facility operation.",
  "canonicalUrl": "https://semiaifoundry.com/learn/atlas/rack-dynamic-power/",
  "publishedDate": null,
  "revisionDate": "2026-10-07",
  "author": null,
  "topic": "Data-center physical systems",
  "series": null,
  "license": null,
  "rightsUrl": "https://semiaifoundry.com/ai-atlas/licenses.html",
  "version": "1.3.0",
  "status": "Published educational lesson",
  "evidenceScope": "No universal AI load waveform is assumed. Real simultaneous behavior and control response need facility and device telemetry.",
  "contentCoverage": "published-lesson",
  "artifactUrl": "https://semiaifoundry.com/ai-atlas/lesson-data/rack-dynamic-power-8c5f087739d666b7.json",
  "interactiveUrl": "https://semiaifoundry.com/ai-atlas/#rack-dynamic-power",
  "related": [],
  "supersedes": [],
  "supersededBy": [],
  "contentHash": "596ded60dbb849d255f332e2d0e37bfa489322d6701c5b5d4bb5e0355c7ed425",
  "contentLength": 4703,
  "jsonUrl": "https://semiaifoundry.com/agents/publications/atlas-lesson-rack-dynamic-power.json",
  "markdownUrl": "https://semiaifoundry.com/agents/publications/atlas-lesson-rack-dynamic-power.md",
  "sections": [
    {
      "id": "section-1",
      "title": "Dynamic load, reserve and power control",
      "start": 45,
      "end": 397
    },
    {
      "id": "section-2",
      "title": "The mechanism",
      "start": 397,
      "end": 1511
    },
    {
      "id": "section-3",
      "title": "Science and assumptions",
      "start": 1511,
      "end": 1539
    },
    {
      "id": "section-4",
      "title": "Energy integral",
      "start": 1539,
      "end": 2115
    },
    {
      "id": "section-5",
      "title": "Engineering example",
      "start": 2115,
      "end": 2323
    },
    {
      "id": "section-6",
      "title": "Distinctions and tradeoffs",
      "start": 2323,
      "end": 2714
    },
    {
      "id": "section-7",
      "title": "Specifications and interfaces",
      "start": 2714,
      "end": 3012
    },
    {
      "id": "section-8",
      "title": "Interfaces",
      "start": 3012,
      "end": 3178
    },
    {
      "id": "section-9",
      "title": "Check your understanding",
      "start": 3178,
      "end": 3207
    },
    {
      "id": "section-10",
      "title": "What additional quantity is needed to turn power into energy?",
      "start": 3207,
      "end": 3496
    },
    {
      "id": "section-11",
      "title": "Evidence and scope",
      "start": 3496,
      "end": 3790
    },
    {
      "id": "section-12",
      "title": "Sources",
      "start": 3790,
      "end": 4246
    },
    {
      "id": "section-13",
      "title": "Continue exploring",
      "start": 4246,
      "end": 4269
    },
    {
      "id": "section-14",
      "title": "Foundations",
      "start": 4269,
      "end": 4473
    },
    {
      "id": "section-15",
      "title": "Related lessons",
      "start": 4473,
      "end": 4703
    }
  ],
  "citations": [
    {
      "title": "Explore in Atlas",
      "url": "https://semiaifoundry.com/ai-atlas/#rack-dynamic-power"
    },
    {
      "title": "Key Questions on Energy and AI: Executive Summary",
      "url": "https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary"
    },
    {
      "title": "NVIDIA DGX GB Rack Scale Systems: Hardware",
      "url": "https://docs.nvidia.com/dgx/dgxgb200-user-guide/hardware.html"
    },
    {
      "title": "Published lesson data",
      "url": "https://semiaifoundry.com/ai-atlas/lesson-data/rack-dynamic-power-8c5f087739d666b7.json"
    },
    {
      "title": "Atlas releases and licensing",
      "url": "https://semiaifoundry.com/ai-atlas/licenses.html"
    },
    {
      "title": "Power supply and power shelf",
      "url": "https://semiaifoundry.com/learn/atlas/power-supply/"
    },
    {
      "title": "Cluster scheduler and resource allocation",
      "url": "https://semiaifoundry.com/learn/atlas/scheduler/"
    },
    {
      "title": "Power integrity and transient voltage margin",
      "url": "https://semiaifoundry.com/learn/atlas/power-integrity-budget/"
    },
    {
      "title": "Uninterruptible power supply and backup chain",
      "url": "https://semiaifoundry.com/learn/atlas/ups/"
    }
  ],
  "recordHash": "9a21bd61818b1e54df95638bdc1d20279c2da82e9bead968c68b202e1eca52bf",
  "contentFormat": "text/markdown",
  "content": "Data-center physical systems · Atlas v1.3.0\n\n# Dynamic load, reserve and power control\n\nAccounts for changes in electrical demand over timescales from chip activity to facility operation.\n\n[Explore in Atlas](https://semiaifoundry.com/ai-atlas/#rack-dynamic-power)Chip voltage rails to facility heat rejection\n\nLearning objectiveMatch power measurements and controls to their relevant timescales.\n\n## The mechanism\n\nAI workloads alternate between arithmetic, communication, memory movement and idle intervals. Their electrical demand changes accordingly, and synchronized jobs can align changes across many devices. Power supplies, local decoupling, rack distribution and facility controls respond on different timescales. A daily energy total cannot reveal a fast voltage event, while a brief peak does not define annual energy. Power capping or workload scheduling can limit demand but may change completion time or service latency. The IEA discusses emerging power-system challenges from AI growth; a local engineering model must still use measured waveforms and the actual facility response.\n\n1.  Record load at the timescale needed to observe the suspected event.\n2.  Separate local voltage dynamics from rack and facility power changes.\n3.  Estimate how many jobs can align their demand and what reserve can respond in time.\n4.  Apply caps or scheduling controls and measure their effect on useful results.\n5.  Integrate power over time for energy accounting without losing peak and recovery information.\n\n## Science and assumptions\n\n### Energy integral\n\n$$\nE=\\int_{t_0}^{t_1}P(t)\\,\\mathrm dt\n$$\n\nEnergy is the integral of power over time at a fixed metering boundary.\n\nWith watts and seconds the result is joules; this card uses MW and hours to obtain MWh.\n\nSymbols and units\n\nE\n\nEnergy consumed over the selected interval\n\nMWh\n\nP(t)\n\nTime-varying real power at the fixed metering boundary\n\nMW\n\nt\n\nTime\n\nh\n\nt₀,t₁\n\nInterval start and end\n\nh\n\nWith P in MW and time in hours, E is MWh.\n\nAssumptions Metering boundary remains fixed; coarse samples can miss peaks even when the integral is approximately correct.\n\n## Engineering example\n\nSynthetic workload draws 1 MW for ten minutes and 0.5 MW for twenty minutes. It uses about 0.333 MWh over the half hour, while its peak remains 1 MW rather than the 0.667 MW average.\n\n## Distinctions and tradeoffs\n\nAn average megawatt figure cannot prove adequate transient response, and a short peak is not a continuous energy demand.\n\nPower limits can improve electrical predictability and capacity sharing while reducing peak throughput or extending job duration.\n\nUseful AI work depends on surviving demand changes without excessive throttling, errors or interruptions.\n\n## Specifications and interfaces\n\nTimescale\n\nFast transient, workload interval and billing period\n\nMeasurements need sufficient temporal resolution for the question.\n\nReserve\n\nAvailable response beyond committed demand\n\nSpecify electrical capacity, energy duration and control latency separately.\n\n### Interfaces\n\nSynchronized workload activity Input\n\nTime-varying current and power requests\n\nPower-control response Output\n\nRegulated supply, caps or shifted work\n\n## Check your understanding\n\n### What additional quantity is needed to turn power into energy?\n\n1.  The duration or time history\n2.  Only the GPU model name\n3.  The number of Ethernet ports\n\nAnswer and reasoning\n\nThe duration or time history\n\nEnergy integrates power over an interval, so both demand and time matter.\n\n## Evidence and scope\n\nNo universal AI load waveform is assumed. Real simultaneous behavior and control response need facility and device telemetry.\n\nNo current calculator directly models this lesson’s main mechanism. Use the library for related bounds while retaining their stated omissions.\n\n## Sources\n\n-   [Key Questions on Energy and AI: Executive Summary](https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary)\n-   [NVIDIA DGX GB Rack Scale Systems: Hardware](https://docs.nvidia.com/dgx/dgxgb200-user-guide/hardware.html)\n\n[Published lesson data](https://semiaifoundry.com/ai-atlas/lesson-data/rack-dynamic-power-8c5f087739d666b7.json) · [Atlas releases and licensing](https://semiaifoundry.com/ai-atlas/licenses.html)\n\n## Continue exploring\n\n### Foundations\n\n-   [Power supply and power shelf](https://semiaifoundry.com/learn/atlas/power-supply/)\n-   [Cluster scheduler and resource allocation](https://semiaifoundry.com/learn/atlas/scheduler/)\n\n### Related lessons\n\n-   [Power integrity and transient voltage margin](https://semiaifoundry.com/learn/atlas/power-integrity-budget/)\n-   [Uninterruptible power supply and backup chain](https://semiaifoundry.com/learn/atlas/ups/)"
}