{
  "schemaVersion": 1,
  "id": "atlas-lesson-optics",
  "title": "Optical transceiver and fiber link",
  "category": "learning",
  "format": "Atlas lesson",
  "description": "Converts electrical communication into light for transport through fiber.",
  "canonicalUrl": "https://semiaifoundry.com/learn/atlas/optics/",
  "publishedDate": null,
  "revisionDate": "2026-10-07",
  "author": null,
  "topic": "Data-center fabric",
  "series": null,
  "license": null,
  "rightsUrl": "https://semiaifoundry.com/ai-atlas/licenses.html",
  "version": "1.3.0",
  "status": "Published educational lesson",
  "evidenceScope": "The lane rates and efficiency are synthetic. Lane/port grouping, modulation, encoding and direction must match the actual module; electrical-to-optical conversion does not remove serialization or protocol overhead.",
  "contentCoverage": "published-lesson",
  "artifactUrl": "https://semiaifoundry.com/ai-atlas/lesson-data/optics-c6affce300ad8ea0.json",
  "interactiveUrl": "https://semiaifoundry.com/ai-atlas/#optics",
  "related": [],
  "supersedes": [],
  "supersededBy": [],
  "contentHash": "5b4b076a7cafecebb607c523201b4cf1e08d869bb22542e1c16ea88cf7635c31",
  "contentLength": 4744,
  "jsonUrl": "https://semiaifoundry.com/agents/publications/atlas-lesson-optics.json",
  "markdownUrl": "https://semiaifoundry.com/agents/publications/atlas-lesson-optics.md",
  "sections": [
    {
      "id": "section-1",
      "title": "Optical transceiver and fiber link",
      "start": 35,
      "end": 389
    },
    {
      "id": "section-2",
      "title": "The mechanism",
      "start": 389,
      "end": 1373
    },
    {
      "id": "section-3",
      "title": "Science and assumptions",
      "start": 1373,
      "end": 1401
    },
    {
      "id": "section-4",
      "title": "Lane aggregation and units",
      "start": 1401,
      "end": 2271
    },
    {
      "id": "section-5",
      "title": "Engineering example",
      "start": 2271,
      "end": 2506
    },
    {
      "id": "section-6",
      "title": "Distinctions and tradeoffs",
      "start": 2506,
      "end": 2885
    },
    {
      "id": "section-7",
      "title": "Specifications and interfaces",
      "start": 2885,
      "end": 3163
    },
    {
      "id": "section-8",
      "title": "Interfaces",
      "start": 3163,
      "end": 3301
    },
    {
      "id": "section-9",
      "title": "Check your understanding",
      "start": 3301,
      "end": 3330
    },
    {
      "id": "section-10",
      "title": "What raw byte rate corresponds to 400 Gb/s before overhead?",
      "start": 3330,
      "end": 3575
    },
    {
      "id": "section-11",
      "title": "Evidence and scope",
      "start": 3575,
      "end": 3967
    },
    {
      "id": "section-12",
      "title": "Sources",
      "start": 3967,
      "end": 4317
    },
    {
      "id": "section-13",
      "title": "Continue exploring",
      "start": 4317,
      "end": 4340
    },
    {
      "id": "section-14",
      "title": "Foundations",
      "start": 4340,
      "end": 4446
    },
    {
      "id": "section-15",
      "title": "Related lessons",
      "start": 4446,
      "end": 4744
    }
  ],
  "citations": [
    {
      "title": "Explore in Atlas",
      "url": "https://semiaifoundry.com/ai-atlas/#optics"
    },
    {
      "title": "NVIDIA — Transceivers and Fiber: 100G-PAM4",
      "url": "https://docs.nvidia.com/networking/display/400g100gpam4ovdev/transceivers-and-fiber-details-100g-pam4"
    },
    {
      "title": "Published lesson data",
      "url": "https://semiaifoundry.com/ai-atlas/lesson-data/optics-c6affce300ad8ea0.json"
    },
    {
      "title": "Atlas releases and licensing",
      "url": "https://semiaifoundry.com/ai-atlas/licenses.html"
    },
    {
      "title": "Network interface adapter",
      "url": "https://semiaifoundry.com/learn/atlas/network-adapter/"
    },
    {
      "title": "Leaf switch and endpoint network",
      "url": "https://semiaifoundry.com/learn/atlas/leaf-switch/"
    },
    {
      "title": "Spine switching and fabric topology",
      "url": "https://semiaifoundry.com/learn/atlas/spine-switch/"
    },
    {
      "title": "NVLink copper cable cartridge",
      "url": "https://semiaifoundry.com/learn/atlas/cable-cartridge/"
    }
  ],
  "recordHash": "f95c14a103268043673ffc9b540adddf74314db867358c38d10393f1a40eca36",
  "contentFormat": "text/markdown",
  "content": "Data-center fabric · Atlas v1.3.0\n\n# Optical transceiver and fiber link\n\nConverts electrical communication into light for transport through fiber.\n\n[Explore in Atlas](https://semiaifoundry.com/ai-atlas/#optics)Centimeter-scale module and meter-to-kilometer links; reach varies\n\nLearning objectiveConvert optical lane rates into a payload byte rate while retaining direction and overhead.\n\n## The mechanism\n\nAn optical transceiver bridges electrical signaling at a network device and light traveling in fiber. Its transmitter modulates optical energy, and the receiver converts the arriving signal back into an electrical form that the system can interpret. Link performance depends on compatible rates, modulation, fiber type, connectors, and reach. A module's headline rate may combine several lanes or ports; an eight-hundred-gigabit module is not automatically one eight-hundred-gigabit optical channel. The atlas separates electrical lanes, optical paths, and the remote receiver so the conversion boundary is clear. It also distinguishes bits per second in networking from bytes per second commonly used for memory bandwidth, preventing an eightfold unit mistake.\n\n1.  Encode the local electrical data onto an optical signal.\n2.  Transport that light through the specified fiber and connectors.\n3.  Detect and recover the remote data within the link’s signal budget.\n\n## Science and assumptions\n\n### Lane aggregation and units\n\n$$\nB_{\\mathrm{bytes}}=\\frac{N_{\\mathrm{lanes}}R_{\\mathrm{lane,bits}}\\eta_{\\mathrm{payload}}}{8}\n$$\n\nAggregate payload bandwidth equals lane count times raw per-lane bit rate times payload efficiency, divided by eight.\n\nSymbols and units\n\nB\\_bytes\n\nAggregate directional payload byte rate\n\nbytes/s\n\nN\\_lanes\n\nIncluded active lane count\n\nlanes\n\nR\\_lane,bits\n\nRaw bit rate per lane\n\nbits/(s·lane)\n\nη\\_payload\n\nFraction of physical bits carrying counted payload\n\nfraction\n\n8\n\nBits per byte\n\nbits/byte\n\nModule rate can aggregate lanes. Dividing bits by eight converts to bytes, and payload efficiency accounts for encoding and protocol overhead rather than claiming every physical bit is application data.\n\nAssumptions Rates must use consistent lane definitions and direction; actual modules may split aggregate capacity into separate ports.\n\n## Engineering example\n\nSynthetic optical interface: eight lanes at 50 Gbit/s each provide 400 Gbit/s raw aggregate rate. At assumed 90% payload efficiency, useful rate is 360 Gbit/s, or 45 GB/s after dividing by eight bits per byte.\n\n## Distinctions and tradeoffs\n\n800 Gb/s is not 800 GB/s, and a module’s aggregate rate may be split across multiple ports or lanes.\n\nOptics supports useful reach and bandwidth density. Conversion power, cost, and compatibility remain design constraints.\n\nFiber connects distant parts of an AI cluster. Reliable links let communication scale beyond short local electrical paths.\n\n## Specifications and interfaces\n\nPublished module example\n\n800G twin-port OSFP\n\nAn example can provide two 400G optical ports.\n\nRate units\n\nGb/s vs GB/s\n\nEight bits equal one byte before protocol overhead.\n\nReach\n\nPart-specific\n\nMust match fiber and connector specifications.\n\n### Interfaces\n\nElectrical transmitter lanes Input\n\nEncoded network data\n\nRemote receiver Output\n\nRecovered data after optical transport\n\n## Check your understanding\n\n### What raw byte rate corresponds to 400 Gb/s before overhead?\n\n1.  400 GB/s\n2.  3,200 GB/s\n3.  50 GB/s\n\nAnswer and reasoning\n\n50 GB/s\n\nDivide the bit rate by eight. Protocol overhead and application behavior reduce usable throughput further.\n\n## Evidence and scope\n\nThe lane rates and efficiency are synthetic. Lane/port grouping, modulation, encoding and direction must match the actual module; electrical-to-optical conversion does not remove serialization or protocol overhead.\n\nUse the execution investigation for the shared tensor-to-memory-to-result thread; no current standalone calculator models this lesson’s full mechanism.\n\n## Sources\n\n-   [NVIDIA — Transceivers and Fiber: 100G-PAM4](https://docs.nvidia.com/networking/display/400g100gpam4ovdev/transceivers-and-fiber-details-100g-pam4)\n\n[Published lesson data](https://semiaifoundry.com/ai-atlas/lesson-data/optics-c6affce300ad8ea0.json) · [Atlas releases and licensing](https://semiaifoundry.com/ai-atlas/licenses.html)\n\n## Continue exploring\n\n### Foundations\n\n-   [Network interface adapter](https://semiaifoundry.com/learn/atlas/network-adapter/)\n\n### Related lessons\n\n-   [Leaf switch and endpoint network](https://semiaifoundry.com/learn/atlas/leaf-switch/)\n-   [Spine switching and fabric topology](https://semiaifoundry.com/learn/atlas/spine-switch/)\n-   [NVLink copper cable cartridge](https://semiaifoundry.com/learn/atlas/cable-cartridge/)"
}