Copper, AEC, and optics: choose the reach boundary before choosing a cable

Reach is an input, not an afterthought

Copper and fiber are not interchangeable versions of the same purchase. Passive direct-attach copper (DAC) is a fixed copper assembly with no active conditioning in the cable. An active electrical cable (AEC), sometimes called an active copper cable, adds electronics in the ends to improve the electrical signal budget. An active optical cable (AOC) performs electrical-to-optical conversion inside a fixed cable assembly. A separate optical transceiver plus fiber separates the removable optics from the cable plant. Each choice changes reach, bend and routing behavior, power, thermal load, serviceability, and the fault-isolation procedure.

A cable’s reach must be read at its exact port speed, gauge, connector, and topology. A number from an older generation cannot be extended to a newer SerDes lane. NVIDIA’s copper overview, for example, describes particular 800G copper and linear-active configurations and lengths; it does not publish a universal copper reach rule. The NVIDIA design guide also explains the basic mechanism: active copper adds signal-boosting electronics in the connector.

The mechanism behind the boundary

Electrical signals lose amplitude and shape as copper gets longer. The host transmitter, cable construction, connector, and receiver together decide whether the receiver can distinguish symbols. An AEC can add retiming or linear amplification depending on its design, but it is still bounded by its specified electrical channel. Fiber moves the long path into an optical channel: transceiver or AOC electronics convert at each end, then the fiber carries light. That can support different reaches and routing patterns, at the cost of optical components, cleaning discipline, and different diagnostics.

Credo’s ZeroFlap product page lists AEC families at several host rates including 800G and 1.6T. This establishes that the vendor markets those categories; use the specific datasheet and approved system list for a deployed length and form factor. A 1.6T AEC listing is not evidence that it fits a particular switch cage, operates at every temperature, or substitutes for fiber across a room.

1. 1same-rack short path
2. 2passive DAC
3. 3lowest component complexity if channel margin permits

1. 1shorter electrical path with loss
2. 2AEC or linear active cable
3. 3verify host and cable compatibility

1. 1row or room path
2. 2AOC or transceiver plus fiber
3. 3validate optics, fiber, and operations

Worked example: a two-row cluster layout

Assume compute racks are placed in two rows and leaf switches are in the middle of each row. Start by measuring—not guessing—the routed path from each NIC to its assigned switch, including vertical routing and slack. Classify the paths into a short in-rack set, a same-row set, and a cross-row set. For each set, record port speed, breakout direction, connector family, required redundancy, airflow constraints, and whether a cable must be field-replaceable independently of optics.

The short in-rack set may be candidates for passive copper if the platform’s validated cable list permits the actual length. The same-row set may have an AEC option. The cross-row set may require an optical design. The conclusion is not “copper loses at distance”; it is that a particular physical path leaves the published electrical envelope. NVIDIA’s ConnectX-8 release notes show how restrictive this can be: they identify a specific 1.6T active-copper cable part and a specific length. That is a useful reminder to treat part number and release support as part of the link contract.

Limits that change the operating model

Copper can be attractive for short links because it can avoid separate optical modules and fiber handling. It can also create bundle weight, bend, and airflow constraints. AEC electronics consume power and require compatibility at both ends. AOC simplifies some installation because cable and optics travel together, while a transceiver plus fiber can make replacement boundaries more modular. None of these choices can be judged from purchase price alone: spare strategy, port density, power delivery, cleaning tools, and time to isolate a fault matter in a dense fabric.

Do not use a vendor’s maximum reach marketing claim as an installation plan. Validate with the intended NIC, switch, firmware, cable revision, FEC mode, and routing condition. A link that trains on a bench can still fail after cable dressing, higher ambient temperature, or a port-mode change.

Lab: draw a physical-link decision sheet

For one row, make a table with source rack, destination rack, routed length, speed, port mode, approved cable or module, optical-fiber type if used, FEC mode, and replacement method. Add an “evidence” column containing the exact compatibility source. Then simulate one failure: which item would be swapped first, and which counters distinguish a cable, module, or host problem? This is an offline planning lab. It does not certify a cable installation.