{"id":"optics","level":"datacenter","name":"Optical transceiver and fiber link","shortName":"Optical link","role":"Converts electrical communication into light for transport through fiber.","description":"An 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.","specs":[{"label":"Published module example","value":"800G twin-port OSFP","context":"An example can provide two 400G optical ports."},{"label":"Rate units","value":"Gb/s vs GB/s","context":"Eight bits equal one byte before protocol overhead."},{"label":"Reach","value":"Part-specific","context":"Must match fiber and connector specifications."}],"interfaces":[{"name":"Electrical transmitter lanes","direction":"in","signal":"Encoded network data"},{"name":"Remote receiver","direction":"out","signal":"Recovered data after optical transport"}],"tradeoff":"Optics supports useful reach and bandwidth density. Conversion power, cost, and compatibility remain design constraints.","whyAI":"Fiber connects distant parts of an AI cluster. Reliable links let communication scale beyond short local electrical paths.","related":["leaf-switch","spine-switch","cable-cartridge"],"sources":[{"title":"NVIDIA — Transceivers and Fiber: 100G-PAM4","url":"https://docs.nvidia.com/networking/display/400g100gpam4ovdev/transceivers-and-fiber-details-100g-pam4"}],"scale":"Centimeter-scale module and meter-to-kilometer links; reach varies","kind":"optics","mechanism":["Encode the local electrical data onto an optical signal.","Transport that light through the specified fiber and connectors.","Detect and recover the remote data within the link’s signal budget."],"misconception":"800 Gb/s is not 800 GB/s, and a module’s aggregate rate may be split across multiple ports or lanes.","check":{"question":"What raw byte rate corresponds to 400 Gb/s before overhead?","options":["400 GB/s","3,200 GB/s","50 GB/s"],"answer":2,"explanation":"Divide the bit rate by eight. Protocol overhead and application behavior reduce usable throughput further."},"branch":"Data-center fabric","subbranch":"Optical transport","science":[{"title":"Lane aggregation and units","equation":"B_bytes = (N_lanes × R_lane,bits × η_payload) / 8","explanation":"Module 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.","assumptions":"Rates must use consistent lane definitions and direction; actual modules may split aggregate capacity into separate ports."}],"history":[],"prerequisites":["network-adapter"],"learningObjective":"Convert optical lane rates into a payload byte rate while retaining direction and overhead.","engineeringExample":"Synthetic 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.","evidenceNotes":"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.","labId":"library","scope":"Use the execution investigation for the shared tensor-to-memory-to-result thread; no current standalone calculator models this lesson’s full mechanism.","investigationId":"execution","labScope":"Use the execution investigation for the shared tensor-to-memory-to-result thread; no current standalone calculator models this lesson’s full mechanism."}