A fiber transceiver is the small pluggable module that lets a switch, router, or server talk over fiber optic cable. It converts electrical signals into light on the transmit side and light back into electrical signals on the receive side. Pick the wrong one and the link either will not come up at all, or it comes up and quietly drops packets under load.
This guide covers the form factors you will actually encounter in 2026 (SFP through QSFP-DD and OSFP), the real distance limits for each, how to read a part number, and the mistakes that cause most failed installs. Our team at Cablify specs and installs these across Toronto and the GTA, so the numbers below reflect what works in the field, not just what the datasheet claims.
What a Fiber Transceiver Actually Does
Inside every module there are two halves. The transmit side uses a laser (VCSEL for multimode, DFB or EML for single-mode) to turn the electrical signal into pulses of light. The receive side uses a photodiode to turn incoming light back into electrical signal. Everything else in the module exists to support those two functions: clock recovery, temperature compensation, and the EEPROM that tells the switch what the module is.
That EEPROM matters more than people expect. It holds the vendor ID, part number, supported speed, and wavelength. Switches read it at insertion and decide whether to accept the module. It also drives Digital Optical Monitoring (DOM), which reports live transmit power, receive power, temperature, and bias current. DOM is the single most useful troubleshooting tool on a fiber link, and it is free once the module is in.
Fiber Transceiver Form Factors Compared
Form factor describes the physical shape and the electrical interface, not the speed on its own. A QSFP28 cage, for example, will take 100G, 50G, and often 40G modules. Here is how the current lineup breaks down.
| Form Factor | Typical Speed | Lanes | Connector | Realistic Distance Range | Status in 2026 |
|---|---|---|---|---|---|
| SFP | 100M – 4.25G | 1 | LC duplex | 550 m MMF to 80 km SMF | Current, widely used |
| SFP+ | 10G (16G FC) | 1 | LC duplex | 300 m OM3 to 80 km SMF | Current, the workhorse |
| SFP28 | 25G | 1 | LC duplex | 70–100 m OM4 to 10 km SMF | Current, growing |
| XFP | 10G | 1 | LC duplex | 300 m to 80 km | Legacy, replaced by SFP+ |
| QSFP | 4G (4 × 1G) | 4 | MPO-12 | Short reach only | Obsolete |
| QSFP+ | 40G | 4 × 10G | MPO-12 or LC | 150 m OM4 to 40 km SMF | Mature, still deployed |
| QSFP28 | 100G | 4 × 25G | MPO-12 or LC | 100 m OM4 to 80 km SMF | Current, dominant at 100G |
| QSFP56 | 200G | 4 × 50G PAM4 | MPO-12 or LC | 100 m OM4 to 10 km SMF | Current, niche |
| QSFP-DD | 400G / 800G | 8 × 50G or 8 × 100G | MPO-12/16 or LC | 100 m OM4 to 40 km SMF | Current standard for 400G |
| OSFP | 400G / 800G / 1.6T | 8 | MPO-16 or LC | 100 m OM4 to 40 km SMF | Current, favoured in AI fabrics |
| CFP / CFP2 / CFP4 | 40G – 400G | Varies | LC duplex | 10 km to coherent long-haul | CFP legacy, CFP2-DCO active |
SFP Transceivers (1G)
The SFP is the original small form-factor pluggable and still the most common module in access-layer switching. It is hot-swappable, single-channel, and uses an LC duplex connector.
- Common variants: 1000BASE-SX (550 m on OM2, 850 nm), 1000BASE-LX (10 km, 1310 nm), 1000BASE-EX (40 km), 1000BASE-ZX (80 km, 1550 nm), 100BASE-FX for legacy fast Ethernet.
- Also supports: 1G/2G/4G Fibre Channel, SONET OC-3 through OC-48, SDH.
- Best for: building-to-building uplinks, IDF-to-MDF runs, camera and access control aggregation, anything where 1G is genuinely enough.
- Watch out for: running 1000BASE-LX over existing multimode fiber. It works, but only with a mode conditioning patch cable to prevent differential mode delay.
SFP+ Transceivers (10G)
Same physical size as an SFP, ten times the throughput. SFP+ is the most widely deployed fiber module in enterprise networks today and it is what most 10G switch ports expect.
- Common variants: 10GBASE-SR (300 m OM3, 400 m OM4), 10GBASE-LR (10 km), 10GBASE-ER (40 km), 10GBASE-ZR (80 km), 10GBASE-LRM (220 m on legacy FDDI-grade multimode).
- Also supports: 8G and 16G Fibre Channel, OTU2.
- Best for: server access, storage networks, campus and data centre uplinks, wireless controller backhaul.
- Cost tip: for runs under 7 m inside a rack, a 10G DAC (direct attach copper) cable is cheaper than two SFP+ modules plus a patch cord, and it draws less power.
SFP28 Transceivers (25G)
SFP28 fits the same cage as SFP+ but runs a single 25G lane. It exists because 25G server NICs deliver most of the benefit of 100G at a fraction of the cost per port, and because breaking a 100G QSFP28 switch port into 4 × 25G is an efficient way to feed a rack.
- Common variants: 25GBASE-SR (70 m OM3, 100 m OM4), 25GBASE-LR (10 km), 25GBASE-ER (40 km).
- Best for: modern server access ports, hyperconverged clusters, 5G fronthaul.
- Watch out for: FEC (forward error correction) settings. 25G links frequently fail to come up because the two ends disagree on RS-FEC, FC-FEC, or no FEC. Check both sides before assuming a bad module.
QSFP+ Transceivers (40G)
QSFP+ carries 40G as four parallel 10G lanes. This is the point where connector type starts to matter: parallel optics use an MPO-12 ribbon connector, while WDM variants multiplex all four lanes onto one fiber pair and use standard LC.
- 40GBASE-SR4: 100 m on OM3, 150 m on OM4, MPO-12 multimode.
- 40GBASE-LR4: 10 km on single-mode, LC duplex, four CWDM wavelengths around 1310 nm.
- 40GBASE-ER4: 40 km on single-mode, LC duplex.
- Breakout option: a single SR4 port splits into 4 × 10G SFP+ with an MPO-to-LC fan-out cable. Useful for getting more 10G ports out of a fixed switch.
- Note on “QSFP”: the original QSFP was 4 × 1G and is effectively extinct. When someone says QSFP today they almost always mean QSFP+ or QSFP28.
QSFP28 Transceivers (100G)
QSFP28 is the volume 100G module: four 25G NRZ lanes in the same cage size as QSFP+. There are more reach variants here than in any other form factor, and picking the right one is largely a fiber-plant question.
- 100GBASE-SR4: 70 m OM3, 100 m OM4, MPO-12 multimode.
- 100G PSM4: 500 m single-mode, MPO-12. Cheaper than CWDM4 but burns eight fibers.
- 100G CWDM4: 2 km single-mode, LC duplex. The sweet spot for campus links.
- 100GBASE-LR4: 10 km single-mode, LC duplex.
- 100GBASE-ER4 / ZR4: 40 km and 80 km respectively.
- Breakout option: 4 × 25G SFP28 from one SR4 port.
QSFP-DD and OSFP (400G and 800G)
Both form factors double the electrical lane count to eight and use PAM4 signalling to get 50G or 100G per lane. QSFP-DD is backward compatible with QSFP28 modules, which makes it the easier upgrade path for existing switching. OSFP is slightly larger with a better thermal design, which is why it dominates in GPU and AI cluster fabrics where modules run hot and never idle.
- 400GBASE-DR4: 500 m single-mode, MPO-12.
- 400GBASE-FR4: 2 km single-mode, LC duplex.
- 400GBASE-LR4: 10 km single-mode, LC duplex.
- 400G ZR / ZR+: coherent optics in a pluggable, reaching 80 km to several hundred km without separate transport gear.
- Power and cooling: these modules draw 12–25 W each. Verify your switch’s per-port power budget and airflow before ordering a full tray of them.
CFP, CFP2 and CFP4 Transceivers
CFP modules are physically much larger than the SFP family, which gives them room for the DSP and optics needed for coherent transmission. The original CFP has largely been replaced in enterprise gear, but CFP2-DCO (digital coherent optics) remains standard in metro and long-haul transport for 100G, 200G, and 400G coherent links.
- Speeds: 40G through 400G depending on generation.
- Standards: 100/400 Gigabit Ethernet, OTU4, STM-256.
- Best for: carrier transport, DWDM systems, inter-city links. Rarely seen in enterprise LANs.
XFP Transceivers (Legacy 10G)
XFP predates SFP+ and does the same job at 10G in a larger package. It is still found in older carrier equipment and some SONET gear, but nothing new is being designed around it. If you are replacing XFP modules, check whether the chassis supports an SFP+ line card instead. Long term, sourcing XFP optics gets harder and more expensive.
- Data rate: up to 11.1 Gbps.
- Standards: 10 Gigabit Ethernet, 10G Fibre Channel, SONET OC-192, OTU2.
- Distance: 300 m multimode to 80 km single-mode.
Single-Mode vs Multimode: Getting This Right First
Before you look at form factors, confirm what fiber is already in the building. The transceiver has to match the glass, and you cannot mix them on the same link.
| Specification | Multimode (OM3 / OM4 / OM5) | Single-Mode (OS2) |
|---|---|---|
| Core diameter | 50 µm | 9 µm |
| Wavelength | 850 nm (VCSEL) | 1310 / 1550 nm (DFB, EML) |
| Practical reach | Up to a few hundred metres | Up to 80 km and beyond |
| Module cost | Lower | Higher |
| Cable cost | Higher per metre | Lower per metre |
| Jacket colour | Aqua (OM3/OM4), lime green (OM5) | Yellow |
| Best fit | Inside a building or data hall | Between buildings, campus, WAN |
The rough rule: if the run stays inside one building and is under 100 m, multimode is usually cheaper overall. Anything crossing between buildings, or any run you might want to upgrade past 100G later, should be single-mode. Single-mode has effectively unlimited bandwidth headroom, so it protects the investment in the cable pathway, which is the expensive part to replace.
How to Choose the Right Fiber Transceiver
Work through these in order. Skipping step one is what causes most of the returns we see.
- Check the switch port and its compatibility list. The cage type (SFP+, QSFP28, QSFP-DD) sets your options. Most vendors publish an approved optics matrix, and many switches will refuse or warn on unrecognised modules.
- Confirm the fiber type already installed. Single-mode or multimode, and which OM grade. If nobody knows, test it before you order anything.
- Measure the actual run length. Not the straight-line distance. Include vertical risers, slack loops at both ends, and pathway detours. Real runs are commonly 30–50% longer than the floor plan suggests.
- Calculate the loss budget. Add connector loss (roughly 0.3–0.75 dB per mated pair), splice loss (about 0.1 dB per fusion splice), and fiber attenuation (roughly 0.4 dB/km at 1310 nm, 3 dB/km at 850 nm). The total must sit inside the module’s link budget with margin to spare.
- Match the speed on both ends. Both modules must be the same standard. A 10GBASE-LR will not talk to a 10GBASE-SR, and a 40GBASE-SR4 will not talk to a 40GBASE-LR4.
- Confirm connector and polarity. LC duplex for serial optics, MPO for parallel. MPO polarity (Type A, B, or C) has to be consistent end to end or the link will not come up.
- Decide on OEM vs compatible. Covered below.
OEM vs Third-Party Compatible Transceivers
Branded optics from Cisco, Juniper, HPE, or Arista typically cost several times what an equivalent coded third-party module costs. The optics themselves often come off the same production lines. The difference is the EEPROM coding and who supports it when something breaks.
- Third-party modules work. A reputable vendor codes the EEPROM for your specific switch model and the port comes up normally.
- Support is the real risk. Some vendors will ask you to swap in branded optics before troubleshooting a link. That is rarely a hard refusal, but it can slow down a ticket.
- Cisco specifics: the
service unsupported-transceivercommand allows non-Cisco optics on IOS platforms, though it does log a warning. - Practical split: branded optics on core and WAN-facing links where a vendor TAC case is likely, compatible optics on access and lab links where the cost saving compounds.
- Buy on warranty, not price alone. A lifetime warranty and a documented DOM implementation matter more than saving another few dollars per module.
Common Installation Mistakes and How to Avoid Them
Most fiber links that fail on day one fail for one of these reasons. None of them are transceiver defects.
- Dirty end faces. This is the number one cause, by a wide margin. A single particle on a 9 µm single-mode core can blind the link. Inspect with a fiber scope and clean with a proper click cleaner before every insertion, including modules straight out of the bag.
- Swapped TX and RX. Easy to do with LC duplex when the polarity clip has been reversed. Symptom is a completely dead link with no receive power showing in DOM.
- Too much light on short runs. Plugging an 80 km ZR module into a 20 m link overloads the receiver. You need an inline attenuator, typically 10–15 dB. Symptom is a link that comes up then throws CRC errors.
- Bend radius violations. Fiber tucked tightly behind a patch panel or cinched under a zip tie loses light. Keep bends above 30 mm radius for standard fiber unless it is rated bend-insensitive.
- MPO polarity mismatch. Type B trunk with Type B patch cords is the usual convention. Mixing types silently breaks parallel optics.
- FEC mismatch at 25G and above. Both ends must agree. Check before replacing hardware.
- Skipping certification testing. An OTDR trace and an insertion loss test at handover give you a baseline. Without one, every future problem turns into guesswork. Our fiber termination and OTDR testing work always includes documented test results.
Reading DOM Output to Diagnose a Link
When a link is flapping or throwing errors, DOM tells you where to look before you touch any hardware. On Cisco IOS the command is show interfaces transceiver detail; on Junos it is show interfaces diagnostics optics. Compare what you see against the module datasheet.
- RX power at or near the floor (often around -40 dBm): no light arriving. Check for a broken fiber, swapped polarity, or a dark far end.
- RX power a few dB below expected: dirty or damaged connector, or a run longer than budgeted.
- RX power above the maximum: receiver overload. Add an attenuator.
- TX power low with rising bias current: the laser is degrading. Replace the module before it fails outright.
- High temperature: airflow problem in the chassis, or a high-power module in a cage without adequate cooling.
Where Fiber Transceivers Are Used
- Data centres: leaf-spine fabrics, storage networks, and top-of-rack to server links. Density and power per port drive the module choice as much as speed does.
- Enterprise campuses: connecting building MDFs and floor IDFs where copper distance limits rule out Ethernet. Single-mode between buildings, multimode within.
- Carrier and metro networks: coherent pluggables now handle links that used to need dedicated transport chassis.
- AI and HPC clusters: 400G and 800G OSFP modules connecting GPU nodes, where east-west traffic dwarfs anything going north-south.
- Building systems: aggregating IP cameras, access control, and building automation across a large site, especially where electrical isolation between buildings is a requirement.
Where the Technology Is Heading
Four things are shaping the next few years of optical module design.
- 800G and 1.6T. Driven almost entirely by AI training clusters. 800G OSFP is shipping in volume, and 1.6T is moving from demo to deployment.
- Silicon photonics. Integrating optical functions onto silicon cuts cost per bit and improves yield at high lane rates.
- Linear pluggable optics and co-packaged optics. Both remove or relocate the DSP to cut power consumption, which has become the real ceiling on switch port density.
- Coherent pluggables going mainstream. 400G ZR and ZR+ let a router drive a metro DWDM link directly, collapsing a layer of equipment out of the network.
Frequently Asked Questions
Can I mix transceiver brands on the same link?
Yes. The two ends only need to run the same optical standard, wavelength, and fiber type. A Cisco 10GBASE-LR will link with a Juniper 10GBASE-LR without issue. Brand matters to the switch reading the EEPROM, not to the light travelling down the fiber.
What is the difference between QSFP+ and QSFP28?
Both use the same physical cage and four lanes. QSFP+ runs each lane at 10G for 40G total; QSFP28 runs each lane at 25G for 100G total. Most QSFP28 switch ports will also accept a QSFP+ module, but the reverse is not true.
Do transceivers need to match on both ends of a fiber link?
They must match on standard, speed, wavelength, and fiber type. They do not need to be the same brand or the same part number. One exception is BiDi modules, which come in matched pairs with opposite transmit and receive wavelengths, so those must be deployed as a set.
How long do fiber transceivers last?
Typical MTBF ratings run into the millions of hours, and most modules outlast the switch they are installed in. Heat is the main enemy. Modules in poorly ventilated cabinets fail noticeably sooner, and rising laser bias current in DOM output is the early warning sign.
Can I use a 10G transceiver in a 1G switch port?
No. An SFP+ module will physically fit an SFP cage but will not negotiate. The reverse works on many platforms: a lot of SFP+ ports accept a 1G SFP and drop to 1G speed, though you should confirm it against the switch’s documentation first.
What is a BiDi transceiver?
A bidirectional module sends and receives on a single strand of fiber using two different wavelengths, commonly 1310 nm one way and 1550 nm the other. It halves the fiber count on a link, which is valuable when you are working with a limited existing strand count and pulling new cable is not practical.
When should I use DAC or AOC instead of transceivers?
Use a DAC for in-rack links under about 7 m: it is cheaper than two modules plus a patch cord and uses less power. Use an AOC for runs of roughly 7 to 100 m where both ends are fixed and you want a sealed, pre-terminated assembly. Use discrete transceivers with structured cabling anywhere the link needs to be patched, re-routed, or tested independently.
Getting the Spec Right the First Time
Choosing a fiber transceiver comes down to four things: the switch port you have, the fiber already in the ground, the real distance, and the loss budget between them. Get those right and the module selection follows almost automatically. Get them wrong and you end up either paying for reach you will never use or troubleshooting a link that was never going to close.
If you are planning a new run, upgrading a backbone, or trying to work out why an existing link keeps flapping, Cablify handles the whole path: fiber cabling, fusion splicing, terminations, OTDR certification, and transceiver selection across Toronto, Mississauga, and the GTA.