800G Optical Module Selection Guide: Form Factors, Reach Codes, DOM and Compatibility

Why 800G selection goes wrong

Choosing an 800G module looks simple on paper. You match the port, order the part, and plug it in. In practice, most failures do not come from the speed at all. They come from three details: the form factor does not fit the cage, the reach code does not match the longest link, and the compatibility check fails so the port stays dark.

All three are easy to settle before you buy, and expensive to discover afterwards. This guide walks through the parameters that matter, then finishes with a checklist you can send straight to a supplier.

1. Form factors: QSFP-DD, OSFP and OSFP224

The name tells you the structure, and the structure tells you how many ports fit and how much heat you can remove.

QSFP-DD stands for double density. It packs two rows of contacts into the QSFP footprint, which gives eight lanes in a small port. The advantage is panel density and backward compatibility with the QSFP ecosystem.

OSFP stands for octal, eight channels. The module is physically larger than QSFP-DD, and that extra volume buys thermal headroom. Above 800G, thermal headroom is usually worth more than port density.

OSFP224 raises the single lane rate to 224G, targeting 1.6T. The number refers to the lane rate, not the total module rate. That is the single most common misreading of the naming.

Port size, thermal headroom and lane count are tied together. Decide how much cooling the switch can give you first, then pick the form factor. Do not do it the other way round.

2. Eight lanes of 100G, or four lanes of 200G

An 800G module can be built as eight 100G lanes or four 200G lanes. The 8x100G route uses mature components and spreads cost across more suppliers, at the price of higher lane count, more internal channels and higher power. The 4x200G route is simpler inside, but demands more bandwidth from the SerDes and the optical devices.

The split follows product generation. At 400G and 800G both routes coexist. Moving towards 1.6T, 200G per lane becomes the default choice.

3. Reach codes: what SR, DR, FR, LR, ER and ZR actually mean

Reach codes are not marketing terms. They are physical medium dependent specifications, and they tell you which fiber to use, how far the link can run, and how much optical budget you have.

Two things to keep in mind when reading that table.

First, the steps between codes are jumps, not a continuous scale. There is no module between short reach and mid reach. A link that sits near a boundary either falls short or costs more than it needs to.

Second, the same reach code can mean different fiber types at different speeds. 400G FR4 and 800G FR8 both reach 2km, but the lane counts differ, so existing cabling may or may not be reusable. Check the actual PMD rather than judging by the name.

In most data centers, links land between DR and FR. If the cabinet layout is not final yet, work backwards from the longest link and buy one reach code for the whole floor. Doing it the other way guarantees a handful of links that do not reach.

4. Breakout: one cage, several links

An 800G port does not have to run at 800G. It can be split into 2x400G or 8x100G, which is what breakout means.

This matters for AI clusters, where rail optimized topologies carry a lot of traffic between GPUs of the same index. Splitting one large port into several smaller links matches that traffic shape, so you do not need to buy low speed switches just to handle a few short connections.

Breakout saves switch ports and part numbers. What it costs you is cabling complexity, and a wrong fiber order is harder to trace than a wrong configuration.

5. Three numbers for the cabinet: power, cooling and panel height

Power per port comes first. Current 800G pluggable modules typically run between 15 and 20 watts per port, while linear drive designs can sit near 10 watts. 1.6T modules are often quoted at 25 to 30 watts. Multiply that by the port count and the scale becomes clear. A switch with 64 ports of 800G spends around a kilowatt just on optics.

Cooling budget comes second. Air cooled chassis have a limit, and the higher the module power, the tighter the inlet temperature limit becomes. If the cold aisle is designed for 27 degrees, the chassis needs inlet margin, or summer afternoons will bring unexplained link speed drops.

Panel height comes third. Fitting 32 ports of 800G into 1U leaves only millimeters between modules, and the air path narrows with it. At that density the mechanical design of the module, including heatsink height and internal thermal paths, affects long term stability more than the optical specification does.

Put power per port multiplied by port count into the cabinet power and cooling plan before anything is signed. If that number is missing, it comes back during acceptance testing as reduced link speed.

6. When copper still makes sense: DAC and AEC

Inside a cabinet or between adjacent cabinets, copper remains the cheapest option.

Passive direct attach cables carry no retiming devices, so power draw is close to zero and cost is low. The trade off is reach, which is limited to roughly one or two meters. Beyond that the eye diagram closes and you need an active electrical cable with retiming.

Active electrical cables now cost close to entry level optics, but they keep an advantage in power and failure rate. If copper works inside the rack, use copper. Do not convert everything to optics just for uniformity.

7. DOM: the five readings that matter

Optical modules are among the few components that report on themselves. Digital optical monitoring reads transmit power, receive power, temperature, bias current and supply voltage.

Two of the five do most of the early warning work. Receive power shows whether the link is losing budget, and bias current shows whether the laser is aging. Temperature and voltage are supporting evidence that the environment has changed.

The value of DOM is not that the data can be read. It is that thresholds are set. Monitoring without thresholds is not monitoring.

8. Compatibility: two gates to pass

The first gate is the coding whitelist. Most switches validate the module identity and accept only their own coding or that of approved vendors. A third party module may simply stay dark, with nothing in the log beyond an unsupported module entry. The fix is either the switch whitelist mechanism or reprogramming the module coding. The first is compliant, the second carries risk, and the requirement belongs in the technical terms before purchase.

The second gate is standards compliance. The module must match the MSA form factor and electrical specification, and its management interface must follow CMIS. These two decide whether the module is recognized at all. A module that does not comply may fit physically, but DOM will not read and alarms will not appear.

The difference between an original module and a third party module is often not optical performance. It is whether the switch agrees to recognize it. That single point drives spare part cost for the next three years, so do not judge on the first quotation alone.

9. A checklist to send to any supplier

Before asking, work out the port quantity. Once capacity, per port speed and available panel height are fixed, cage count and patch cord count follow, and procurement has a baseline.

One last item is easy to overlook. Write batch coding consistency into the contract. If the coding rules change between batches of the same part number, you end up with a switch that recognizes half the modules and rejects the rest, and the problem to solve is the purchase ledger rather than the optical path.

Accelink and Eoptolink 800G modules, available through APEX

Two manufacturers account for a large share of the 800G market in volume: Accelink and Eoptolink. Both build the form factors described above in QSFP-DD and OSFP, across the common reach codes from DR through LR and ER, and both are widely deployed in AI data center and carrier networks.

APEX supplies modules from these two brands alongside our own optical transport and synchronization products. The manufacturer is stated on every quotation, so the origin of each part is clear before an order is placed.

If you are already standardized on Accelink or Eoptolink part numbers, send us the list. We will confirm the firmware version, the coding and the lead time, and keep the batch coding consistent for future orders so a mixed fleet does not appear six months later.

Working with APEX

APEX supplies optical modules from 1.25G access optics to 800G and coherent 400G/800G, together with optical transport platforms and network synchronization equipment. Third party brand optics are supplied through APEX, so compatible parts and our own products can be sourced from one place.

Send us the switch model, the reach and the port types, or simply a BOM, and we will come back with the modules that fit, the compatibility position and a delivery date.