SFP vs SFP+ vs SFP28 vs QSFP: transceiver form factors explained

SFP, SFP+, and SFP28 modules are physically interchangeable in a parts bin, yet they follow different electrical specifications, top out at different speeds, and interoperate in only one direction. This guide maps the complete transceiver form-factor ladder — speeds, backward-compatibility rules, breakout options, and cabling — using figures from vendor data sheets and the underlying SFF/IEEE specifications.

The transceiver form-factor ladder

Every pluggable form factor is defined by two things: the mechanical/electrical specification (an SFF or MSA document) that fixes the cage, connector, and lane count, and the IEEE 802.3 physical-layer standards that define what actually travels over the wire or fiber. SFP-family modules carry a single electrical lane; QSFP ("quad") modules carry four. Speed increases came from raising the per-lane signaling rate — 1.25G, then 10.3125G, then 25.78125G NRZ — and, in the 200G/400G generation, from switching to PAM4 modulation and doubling lanes.

Note the pattern in the SR columns: reach does not increase with speed. A 25G or 100G short-reach link on OM3 covers roughly a quarter of the distance a 10G link does, because the faster lane rate is less tolerant of modal dispersion. Budget fiber plant accordingly when upgrading in place.

SFP, SFP+, and SFP28: one cage, three speed generations

The SFP outline defined in the INF-8074i MSA has survived three speed generations essentially unchanged. What differs is the electrical channel behind the 20-pin connector: SFF-8431 qualified it for 10 Gbps signaling (SFP+), and SFF-8402 qualified the same mechanical envelope for the 25.78 Gbps lane rate used by IEEE 802.3by 25 GbE (SFP28). Because the cage and latch are identical, switch vendors build one physical port and rate-limit it in firmware.

• SFP (1G): 1000BASE-SX/LX optics with LC duplex connectors, plus 1000BASE-T copper modules for RJ45 runs. Still common on management networks and campus access gear.
• SFP+ (10G): the enterprise workhorse. A 10GBASE-SR SFP+ transceiver reaches 300 m on OM3 and 400 m on OM4 multimode over LC duplex, per Cisco's 10GBASE SFP+ data sheet; 10GBASE-LR covers 10 km on single-mode.
• SFP28 (25G): one lane borrowed from the 100G electrical spec (IEEE 802.3bj lane technology, standardized for 25 GbE in 802.3by). 25GBASE-SR reaches 70 m on OM3 and 100 m on OM4, and Cisco's data sheet notes the SR optic requires RS-FEC on the host port — FEC configuration is a real operational difference versus 10G.

Because all three generations share a cage, the practical questions are about coding and platform support rather than physical fit — see our SFP transceiver compatibility guide for how vendor coding, third-party optics, and platform allowlists interact.

QSFP+, QSFP28, and the 200G/400G generation

QSFP modules are wider than SFP-family modules because they carry four electrical lanes side by side. QSFP+ (SFF-8436) bundles 4 × 10G lanes into a 40 GbE port; QSFP28 (SFF-8665) bundles 4 × 25G lanes into 100 GbE. The four-lane design is also what makes breakout cabling possible — each lane can be split out to a separate lower-speed port.

On multimode fiber, the four lanes run in parallel over separate fiber strands: a 40GBASE-SR4 QSFP+ transceiver drives 4 × 10G over an MPO-12 trunk to 100 m on OM3 or 150 m on OM4; 100GBASE-SR4 does 4 × 25G over the same MPO-12 plant to 70 m on OM3 or 100 m on OM4. On single-mode, LR4 and CWDM4 optics instead multiplex four wavelengths onto one fiber pair with a standard LC duplex connector.

Beyond 100G, three form factors matter, briefly:

• QSFP56: same QSFP footprint, but each lane moves to 50 Gbps PAM4 signaling for 200 GbE.
• QSFP-DD: "double density" — an eight-lane electrical interface in a QSFP-compatible footprint, giving 400 GbE at 50G PAM4 per lane (and 800G at 100G per lane). QSFP-DD ports are designed to accept older QSFP28/QSFP+ modules, which is a major reason it won the 400G density battle in switching.
• OSFP: a slightly larger eight-lane form factor with more thermal headroom, favored in 400G/800G AI and HPC fabrics. It is not mechanically compatible with QSFP cages.

Backward compatibility: what fits where

The rule of thumb: a newer, faster port usually accepts an older, slower module of the same physical family and runs it at the module's native speed. The reverse never works, because the older host electronics cannot drive the faster lane rate. Cisco's Nexus 9000 hardware guides state this explicitly — SFP28 downlink ports support 25G SFP28, 10G SFP+, and 1G SFP optics.

Three caveats before you rely on this matrix. First, "usually" is doing real work: some platforms set speed per group of ports rather than per port, and some uplink modules are locked to a single rate — check the hardware installation guide. Second, an SFP-family module never fits a QSFP cage directly; some vendors offer QSFP-to-SFP+ adapters (QSA) as a workaround. Third, platform allowlists still apply on top of physical fit — a supported form factor does not guarantee a supported part number. Our Catalyst 9300 transceiver compatibility matrix is a worked example of checking a specific platform before buying.

Breakouts: 40G to 4×10G and 100G to 4×25G

Because QSFP+ and QSFP28 are internally four independent lanes, one QSFP port can be split into four SFP-family links. This is the standard way to fan a spine or top-of-rack switch out to lower-speed servers, and it is often the cheapest path to high port density.

Two prerequisites are frequently missed. The parent switch must support breakout configuration on that specific port (on NX-OS, interface breakout; many platforms restrict which ports can split), and the far end must be four independent ports — a breakout cable cannot merge four 10G ports into one 40G link. Cisco's 40GBASE data sheet also confirms SR4 optics interoperate with 10GBASE-SR in 4×10G breakout mode at full SR4 distances.

DAC vs AOC vs optical modules

Within a form factor, the same port can be cabled three ways, and the right choice is almost entirely a function of distance and budget.

• Passive DAC (direct-attach copper): a twinax cable with the module ends permanently attached and no active electronics. Cisco's 10G passive twinax runs 1–5 m, with active versions at 7 and 10 m; 25G passive DACs also span 1–5 m. Lowest cost, near-zero power draw, and negligible added latency — the default for in-rack server connections. At 25G, FEC rules depend on length: per Cisco's SFP28 data sheet, cables up to 2 m need no FEC, 2.5–3 m need BASE-R FEC, and 4–5 m need RS-FEC on the host ports.
• AOC (active optical cable): a fixed fiber assembly with transceivers bonded to each end. Cisco's 25G AOCs come in 1–10 m lengths. Thinner and easier to route than twinax, immune to the reach cliff, but a failure means replacing the entire assembly, and you cannot re-terminate or patch it.
• Optical module + structured fiber: separate transceivers and field-replaceable jumpers or trunk cabling. Required for anything through a patch panel, across rows, or beyond ~10 m — up to 400 m on multimode at 10G and 10 km or more on single-mode. Highest per-link cost, maximum flexibility.

A practical sequencing rule for a rack build: DAC inside the rack, AOC to the adjacent rack, optics plus structured cabling for everything else.

Connectors and media: LC duplex vs MPO-12

Connector type follows lane architecture, not speed. Serial optics — everything in the SFP family, plus QSFP LR4/CWDM4 optics that multiplex four wavelengths onto one fiber pair — use an LC duplex connector with two fiber strands: one transmit, one receive.

Parallel SR4 optics (40GBASE-SR4, 100GBASE-SR4) instead use an MPO-12 connector carrying twelve fibers, of which eight are active — four transmit and four receive — and four are dark. Cisco's data sheets specify female MPO/MTP connectors on the cabling that mates with SR4 modules. Practical consequences for your fiber plant:

• Existing LC-terminated OM3/OM4 plant cannot carry SR4 links without re-termination or cassettes; the glass is the same, the termination is not.
• MPO polarity must be managed end to end — parallel optics links are commonly built with Type-B (crossed) trunks; a polarity mismatch presents as no link with healthy light levels.
• Moving 40G/100G between buildings usually means switching to LR4/CWDM4 optics on single-mode with LC duplex, which conveniently reuses duplex patch practices.

Wavelengths follow the usual split: 850 nm VCSELs on multimode for SR/SR4 optics, 1310 nm on single-mode for LR-class optics rated to 10 km.

Sourcing transceivers on the secondary market

Transceivers hold up well as refurbished purchases: they have no moving parts, and laser degradation is directly observable — a module reporting healthy TX power and RX sensitivity through digital diagnostics (DOM, per SFF-8472) has verifiable remaining life. When buying used, confirm three things: the vendor coding matches your platform's allowlist, DOM is supported and readable on your switch, and the seller light-tests modules rather than shipping them untested.

The economics favor buyers right now. 40G QSFP+ hardware is heavily discounted as hyperscalers complete 100G/400G migrations, while SFP+ remains the highest-volume enterprise interface and shows no sign of retirement. Whether you need a single spare optic or DAC/AOC cabling for a full rack build, start with our networking catalog and match part numbers against your platform's compatibility matrix before ordering.

Related reading: SFP Transceiver Compatibility Guide · Catalyst 9300 Transceiver Compatibility Matrix

Frequently asked questions

Can I plug an SFP+ transceiver into an SFP28 port?

On most platforms, yes — the port runs at 10G. Cisco Nexus 9000 hardware guides state SFP28 downlink ports support 25G SFP28, 10G SFP+, and 1G SFP optics. Check your platform's hardware guide, since some switches set speed per group of ports rather than per individual port.

Will a 25G SFP28 module work in a 10G SFP+ port?

Not at 25G — the SFP+ host electronics cannot drive the 25.78 Gbps lane rate, and a 25G-only optic generally will not link at all. The exception is dual-rate 10/25G optics (such as Cisco's SFP-10/25G-CSR-S), which are explicitly designed to operate at 10G in SFP+ ports.

Is QSFP28 backward compatible with QSFP+?

In one direction. A QSFP28 port on most switches accepts a QSFP+ module and runs the link at 40G. The reverse fails: a QSFP28 module in a QSFP+ port does not work, because the older port cannot support 25G-per-lane signaling.

How does a 100G to 4x25G breakout work?

Either a QSFP28-to-4xSFP28 passive copper breakout cable (Cisco QSFP-4SFP25G-CU, 1-5 m) or a 100GBASE-SR4 optic feeding an MPO-12 harness that splits into four LC duplex pairs, each landing on a 25GBASE-SR optic. The parent switch must support breakout configuration on that port, and the four child links appear as separate logical interfaces.

When should I use a DAC instead of an optical module?

Inside a rack or to an adjacent rack, at 5 m or less. Passive DACs are the cheapest option, draw almost no power, and add negligible latency. At 25G, watch FEC requirements: per Cisco's data sheet, DACs up to 2 m need no FEC, 2.5-3 m need BASE-R FEC, and 4-5 m need RS-FEC. Beyond 5 m, use an AOC (to about 10 m) or optics with structured fiber.

Do 40G and 100G SR4 optics use the same fiber as 10GBASE-SR?

The same OM3/OM4 multimode glass, but different termination. 10GBASE-SR uses an LC duplex connector over two fibers; SR4 optics require an MPO-12 connector with eight active fibers (four transmit, four receive). Existing LC-terminated plant needs re-termination or MPO cassettes to carry SR4 links.