Enter for the full results page · Esc to close
Modules, Packaging & Form Factors
A pluggable transceiver with the DSP removed — the host SerDes drives the optics directly through a linear analogue path.
A conventional high-speed transceiver contains a digital signal processor. The DSP receives the electrical signal from the switch, recovers the clock, equalises the waveform, re-times it and drives the laser from a clean regenerated copy. It does the same in reverse on the way back. That chip is the single largest consumer of power in the module and it adds latency in both directions.
Linear pluggable optics deletes it. The host SerDes drives the optical front end through a purely analogue, linear amplification path — no retiming, no digital regeneration. The switch ASIC's own equalisation is expected to do the work that the module's DSP used to do.
It is worth being precise about what ‘linear’ means here. The module still contains active electronics: a transimpedance amplifier on receive and a modulator driver on transmit. What it does not contain is anything that makes a decision about what the bits are. The signal that leaves the switch is amplified and shaped, never re-created.
Inside an LPO module you find a linear transimpedance amplifier on the receive side and a linear driver on the transmit side, and very little else in the signal path. Because nothing regenerates the waveform, whatever distortion the host introduces arrives at the laser, and whatever distortion the fibre and the photodiode introduce arrives back at the host SerDes.
That is the whole bargain. You save the DSP's power — commonly 30 to 50 per cent of module power — and you save its latency, because the signal no longer makes a digital round trip. In exchange, the signal-integrity budget now has to be closed across the host, the module and the fibre together, rather than being isolated inside the module.
Forward error correction moves too. In a DSP module the FEC encode and decode usually happen in the module; in an LPO link the host does it. That is not a problem in itself, but it means the host has to support the FEC mode the link needs, which is another thing to check rather than assume.
The DSP is not only a power consumer. It is an isolation layer, and isolation is what makes pluggable optics an open market. Because the DSP cleans up whatever arrives, a compliant module works between a host and a fibre plant it has never met, from vendors who have never spoken to each other.
Remove it and that guarantee weakens. The module's performance now depends on the electrical characteristics of the specific switch board it is plugged into: trace length, connector quality, crosstalk from neighbouring lanes. Two switches that both meet the same specification can present measurably different channels, and an LPO module that is comfortable in one may be marginal in the other.
In practice this is why LPO deployments so far are single-vendor or jointly qualified rather than open-market. It is a systems problem, not a manufacturing one, and it will ease as host SerDes improve and interoperability testing matures — but it is real today and it should shape how you buy.
LPO makes sense where the electrical channel is short and well characterised and where the optical span is modest: server-to-leaf inside a rack, leaf-to-spine within a row. At 800G and above, in an AI cluster where a switch may be dissipating a kilowatt in transceivers alone, removing half the module power is a material result rather than an optimisation.
Work out what that is actually worth to you before deciding. Eight watts saved across 32 ports, over three years, at a realistic electricity price and PUE, is a number the TCO calculator will give you in a few seconds — and it is usually large enough to justify the qualification effort in a homogeneous fabric, and not large enough in a mixed one.
It makes much less sense across a long structured-cabling channel, or in any deployment where the two ends come from different vendors and nobody has jointly qualified the electrical path.
Three questions settle most of it. Which switch platforms has this module been qualified against, and can I see the results? What happens at the far end — is it another LPO module, or a conventional DSP one, and has that combination been tested? And which FEC mode does the host have to run?
A supplier who can answer all three is selling you an engineered link. One who answers ‘it is MSA-compliant’ is selling you a component and leaving the systems risk with you.
| Module power, DSP version | ≈ 16 W |
|---|---|
| Module power, LPO version | ≈ 8–11 W |
| Latency removed | tens of nanoseconds each direction |
| Practical reach | in-rack and in-row; short-reach classes only |
| Interoperability | host and module qualified together |
| Retimer / CDR | none — analogue path end to end |
| FEC | handled by the host, not the module |
LRO keeps a DSP on the transmit side only, which is a much easier thing to deploy across vendors. CPO attacks the same power problem from the opposite end — keep the DSP, delete the circuit board. NPO sits between them, and our four-way comparison works through when each one wins.
Send the specification and an engineer answers it — no obligation, and no sales sequence. If the answer is that you do not need what we sell, that is the answer you get.
Usually answered within one working day.