Frequently asked questions

SFP tops out at 2.5 Gb/s and typically uses FP or DFB lasers with a PIN receiver. SFP+ carries 10 Gb/s in the same physical size and uses higher-linearity optics with stricter electrical specs (SFF-8431). They share the same cage, so an SFP+ port usually accepts an SFP module, but not the reverse.

Choose BiDi when fibre count is the constraint — for example a single-strand drop cable to a remote cabinet, or an access ring where pulling new fibre is expensive. BiDi modules must be paired (1310/1550 with 1550/1310; 1490/1550 with 1550/1490), so order them as an A/B pair.

Longer reaches need a narrower laser linewidth, higher launch power and a more sensitive receiver — usually a cooled DFB or EML with an APD. The optical budget on 80 km spans is also tighter, so the module must hold its specs over temperature. That test and component cost is what you pay for.

No. CWDM channels are 20 nm apart (1270–1610 nm), DWDM channels are 0.8 nm apart in the C-band, and the two grids do not line up. Mixing them will also defeat any EDFA amplifier, which is designed for DWDM channel spacing.

Yes. We programme the EEPROM to the profile your host expects — vendor name, part number and checksum — before shipping. Tell us the switch or router model at the quotation stage and we will match it; unprogrammed generic modules are frequently rejected by locked platforms.

Standard reach codes are usually available from stock or within 2–3 weeks. BiDi, CWDM and industrial-temperature variants take 3–4 weeks because of the wavelength-specific laser. Samples for qualification ship faster — tell us the span and host so we can pick the right code the first time.

Because 25G maps cleanly onto 100G switch ports: 4×25G breakout uses the same QSFP28 port you would otherwise fill with one 100G uplink. Per-gigabit cost of SFP28 is close to SFP+, while switch port density improves fourfold compared with 10G.

A 100G QSFP28-to-4×SFP28 breakout DAC under 3 m, or a breakout AOC for longer runs. The switch must support breakout configuration on that port; check the port profile before ordering, because some platforms ship with all ports in 1×100G mode.

Extended SR (eSR) is a multi-source agreement for 300 m over OM3 rather than the 100 m of standard SR. It is widely supported but not part of IEEE 802.3by, so specify it only when you control both ends of the link.

About 6.3 dB over 10 km of SMF, which leaves comfortable margin for connectors and splices. If your span measures above 8 dB of loss, move to a 25G ER code rather than accepting a marginal link.

The cage is the same, but the port must support 25G signalling. An SFP+ 10G port will not link at 25G. Conversely an SFP28 port usually accepts a 10G SFP+ module, which makes staged 10G-to-25G upgrades practical.

25G SR typically requires ≥ 12 dB return loss on multimode connectors, which means APC is not needed but contaminated or low-grade connectors are. Clean and inspect every end before turn-up — most 25G SR link failures at installation are connector-related, not module-related.

CWDM4 is the right answer at 2 km: it uses the same four wavelengths as LR4 but with lower launch power and a smaller budget, so it costs less. Choose LR4 when the span is genuinely 10 km or when you need a single part number across mixed reach classes.

SR4 is 100 m over OM4 (70 m over OM3). eSR4 extends that to 300 m over OM3/OM4 using a tighter optical design. For newer build-outs with OM4, standard SR4 is enough; eSR4 matters in older OM3-heavy campuses.

Typical QSFP28 dissipation is 3.5–4.5 W and 100G ports are usually densely packed, so airflow matters more than per-module power. Keep the switch's blanking panels fitted and never operate a high-density line card with missing fans or filters.

Yes on most platforms, using a QSFP28-to-4×SFP28 breakout DAC or AOC with the port configured for breakout. This is the standard way to connect 25G servers to a 100G leaf switch.

4WDM-10 is an industrial-temperature 100G 10 km code with a reduced wavelength set, built for outdoor and unheated environments. It trades some budget for a −40 to +85 °C rating, so use it only where temperature is the constraint.

Usually loss and cleanliness: a rack path adds patch panels, more connectors and tighter bend radii. Measure each span end-to-end with an OTDR or light source before turning up, and clean every connector — 100G PAM4 links degrade far earlier than 10G links under the same contamination.

An AOC is a factory-terminated fibre assembly with transceiver-style ends, so no polishing or patch panels are needed. Beyond about 5 m it beats copper DAC on error rate and power, typically covering 1–100 m at 10G–100G.

Within roughly 3–5 m, DAC is cheaper and draws less power. Beyond that DAC becomes thick and stiff while AOC stays slim and reaches 100 m, so 10–30 m row links usually go AOC.

Functionally yes — the lasers and receivers are built into the cable ends, which removes connector loss and cleaning steps. The assembly is not field-repairable, so keep spares for critical links.

10G SFP+ through 100G QSFP28, including 40G QSFP+ and 4x breakout assemblies; every cable is bit-error tested before shipment.

XGPON (ITU-T G.987) delivers 2.5G/2.5G asymmetric; XGS-PON (G.9807.1) is 10G symmetric. Both use 1577/1270 nm, and coexistence with GPON relies on a combo WDM filter.

Usually not — Class B+/C+ splitters are wavelength-agnostic. You replace the OLT port and ONUs, adding a coexistence filter only if GPON must keep running.

20 km class with PR30/PR40 power budgets, SFP OLT and ONU, SC/UPC or SFF receptacles, plus industrial-temperature ONU options.

Yes — with a combo (WDM) OLT module or an external coexistence filter the two systems share one ODN while staying on separate wavelength bands.

OSFP is taller and has a larger top surface, so it dissipates heat more effectively at 12–16 W per module. It also became the preferred form factor for AI fabrics. QSFP-DD800 exists, but high-density 800G deployments overwhelmingly standardise on OSFP.

SR8 is 850 nm VCSEL over multimode, 60 m on OM3 and 100 m on OM4, MPO connectors. DR8 is 1310 nm parallel single-mode to 500 m, dual MPO-12. 2×FR4 uses two CWDM duplex LC pairs up to 2 km. Reach and fibre type decide it, not price.

Budget 12 W for SR8 and up to 16 W for DR8/2×FR4. On a 32-port 800G line card that is roughly 400–500 W of optics alone, so switch airflow, heatsink design and rack power must be planned together — not added afterwards.

Yes on platforms that support breakout, typically with a 2×DR4 or 2×FR4 style module feeding two 400G downstream links. Confirm the switch supports the specific branch mode and that both ends use compatible FEC settings.

Yes. 800G PAM4 relies on RS(544,514) FEC and the link will not meet its BER target without it. The same FEC setting must be used at both ends; a mismatch shows up as a link that trains but never comes up clean.

Mechanically the cage is the same family, but the port must support 8×100G PAM4 and CMIS 5.0 management. Many 400G OSFP ports accept 400G modules only. Verify against the switch's transceiver support matrix before ordering.

400ZR is the OIF implementation agreement: 400G DP-16QAM, 80 km amplified, one fixed application code. 400ZR+ follows the OpenZR+ MSA and adds 100G/200G/300G line rates, longer reach and more FEC options. If your spans vary, ZR+ gives a single part number across them.

OIF specifies 80 km with amplification and roughly 40 km unamplified; real deployments commonly reach 120 km with ZR+ and adequate OSNR. The limit is OSNR and dispersion, not fibre length alone — measure the span before assuming a distance.

For a plain 400GbE client you do not: the module takes 400GbE on the client side and puts a 400G coherent wavelength on the line side. You still need a DWDM multiplexer and, for spans above the unamplified reach, an EDFA line system.

Because the DSP that performs coherent detection and equalisation is the dominant consumer — far more than the laser. 400ZR codes typically run 18–23 W and ZR+ high-power versions up to about 23.5 W. That is why coherent ports are often limited per line card.

One part number covers the whole C-band (1528.77–1566.52 nm) on 50/75/100 GHz grids, so you carry one SKU instead of dozens of fixed-wavelength parts, and you can re-tune a link to a different channel during maintenance without swapping hardware.

Compatibility depends on grid, OSNR and amplifier type rather than the module alone. Open line systems built for OpenZR+ generally interoperate; closed vendor systems may require their own coded optics. Share your line system details and we will confirm before you commit.

A photodiode — PIN for speed and cost, APD for sensitivity — plus a TIA in one package: the receive engine of an optical module.

PIN covers short and metro reaches at lower cost; APD adds 8–12 dB sensitivity for 40 km+ links and GPON Class B+/C+ ONU receivers.

155 Mb/s to 25 Gb/s: 2.5G/10G PIN-TIA for access and metro, 25G PIN-TIA for SFP28, and APD-TIA variants for long reach.

Both — pigtail versions for solder-in module engines, and SC/FC/LC receptacle versions for field-replaceable designs.

It is a metal can with a clear or lensed window that houses photodiodes and some lasers — common in ROSA builds and 155M–2.5G transmitters.

The larger TO-56 can fits bigger lenses and higher-power lasers (10G DFB class); TO-46 suits photodiodes and compact 2.5G-class transmitters. Match your existing BOM footprint.

Yes — PD+TIA matched pairs can ship with responsivity, dark current and transition-time data, so ROSA assembly starts from verified parts.

Yes — window material, lens type and wavelength band are configurable; MOQ applies and samples ship before volume.

Not directly. The shells and pinouts are unrelated, so the SFP has to sit in a purpose-built adapter, and the adapter has to be supported by the installed network OS. Even then the port runs at 10G, so cheaper alternatives usually exist.

No. QSFP is a four-lane envelope, and the oldest version carried 4 × 1G — slower than a single 10G SFP+. Aggregate rate comes from lane count multiplied by lane speed, so a 40G QSFP+ is four lanes at 10G and a 100G QSFP28 is four lanes at 25G.

In most cases yes, and the port steps down to 40G. It is worth confirming against the platform's transceiver matrix, since some operating system releases restrict mixed-rate operation on specific line cards.

It needs a purpose-built assembly: a breakout DAC, a breakout AOC, or a parallel module with an MPO fan-out harness. A plain duplex patch cord cannot carry four lanes, because it has only two fibers.

Two usual causes. The port was never switched into breakout mode in software, so the sub-interfaces exist without a valid lane map; or the server-side optic and the QSFP sub-port disagree on FEC framing. Check the interface configuration first, then the FEC settings at both ends.

They report the same five measurements — case temperature, supply voltage, laser bias, transmit power, receive power — but through different memory maps. That is why a diagnostic script written for an SFP page layout will return nonsense on a QSFP until the map is handled correctly.

Usually yes. Most SFP+ and SFP28 ports negotiate down and will run a 1G module at 1G, provided the port configuration allows it. The reverse never works: a 1G-only cage has a SerDes fixed at 1.25 Gbaud and cannot carry a 10G signal.

They share the same shell and 20-pin connector, so they fit the same opening. They are not interchangeable electrically. The host port must support the module's rate, and the cage must be rated for the signal integrity and heat that rate produces.

An SFP port is a cage that accepts a pluggable module and therefore supports fiber (or copper through a 1000BASE-T SFP). An RJ45 port has the jack built into the chassis and is limited to twisted-pair copper at the speed the PHY supports.

Almost always a coding mismatch rather than a hardware fault. The switch read page A0h, did not recognize the vendor or part identifier, and disabled the port. Confirm the requested coding profile against the switch platform and firmware before assuming the module failed.

Read the diagnostics first. Zero received power with normal bias current points to a broken or disconnected path. Low but non-zero receive power points to loss — dirty connectors, tight bends, too many splices. Received power above the receiver's maximum points to over-driving, which calls for an attenuator. Bias current drifting upward with falling output power is the signature of an ageing laser.

No. Reach class is a budget, not a guarantee. Long-reach parts have higher launch power and tighter receiver specs, and on short spans they can saturate the receiver. Match the class to the measured distance, then verify the budget.