Every pluggable optic in a modern network belongs to a form-factor family, and the family decides what the module can do before any datasheet is opened. SFP, SFP+ and QSFP28 are the three families most buyers meet first — in our own optical transceiver range and in the adapter pages of this site. This guide explains what each one is, what changed between them, and the compatibility rules that follow.
SFP: the gigabit original
SFP (Small Form-factor Pluggable) standardised the idea that the optical interface is a replaceable module. An SFP module carries a transmitter, a receiver and a small identity memory, and slides into a cage on a switch or network card. Its electrical interface is rated for gigabit-class signalling — 1.25G is the classic Ethernet rate — and the family covers single-mode, multimode, copper and BiDi variants. Decades after introduction, SFP remains the connective tissue of management networks, access layers and fiber-to-the-desk, because gigabit is still what those links need.
SFP+: the same shell, ten times the signalling
SFP+ kept the SFP shell and cage dimensions and re-engineered the electrical interface for 10G. The trick of the design is that most of the signal conditioning moved into the host, keeping the module simple and inexpensive — which is why 10G optics became affordable enough to wire entire data centres. An SFP+ module will physically fit an SFP cage, and an SFP module will fit an SFP+ cage; what happens electrically is a separate question, answered below.
QSFP28: four lanes in a wider shell
QSFP28 is a different, wider form factor. "Quad" is literal: four independent lanes, each running at up to 25G, aggregated to 100G in a single module. The cage is larger, the power budget is larger, and the connector at the fiber end is either duplex LC (for single-mode variants) or MPO (for parallel multimode variants). QSFP28 is the module class behind 100G adapters such as the FF-1102E in our 100G network cards, and it also serves as a breakout source: one QSFP28 port can fan out to four 25G SFP28 links.
The compatibility rules that matter
- SFP module in an SFP+ cage: usually works — the host negotiates the module’s gigabit rate. This is the standard way to light a legacy link from a modern switch.
- SFP+ module in an SFP cage: does not give you 10G — the cage’s electronics are gigabit-class. The module may not link at all.
- SFP28 (25G) modules need SFP28 or better hosts; SFP28 cages accept SFP+ modules for staged migrations.
- QSFP28 shares nothing mechanically with the SFP family — different cage, different shell. Adapters between families are not a thing; breakout cables from QSFP28 to SFP28 are.
- In every family, the two ends of a link must match in wavelength and fiber discipline. Form factor compatibility never overrides optics compatibility.
How to choose
Work backwards from the host. The cage on your switch or NIC is fixed; the rate class follows the cage; the variant follows your fiber. If you are buying hosts, buy the cage generation your fabric plan assumes for the next five years — cages outlive the modules that first populate them. If you are buying modules for existing hosts, the decision is already made: match the family, then spend your attention on wavelength, reach and coding.
Coding deserves the last word, because it is where form-factor discussions usually end in the field. Each module carries identity data that the host reads; hosts vary in how strictly they check it. Our compatibility guide explains how coding works and how module-and-host compatibility is established, so a link that should work on paper also works in the rack.
Can I plug a QSFP28 module into an SFP+ cage with an adapter?
No. The families are mechanically and electrically incompatible. Use breakout from a QSFP28 port to SFP28 tails, or specify the right cage on the host.
Is SFP28 the same as SFP+?
Same shell lineage, different electrical generation: SFP28 is the 25G class and supersedes SFP+ in new designs, while accepting SFP+ modules in its cages.
Power and thermals: the quiet difference
Form factor generations also differ in appetite. A gigabit SFP module sips power — typically around a watt. SFP+ modules run a little warmer; QSFP28 modules, driving four lanes and their signal conditioning, are the hungriest of the three and the first to matter in a fully populated chassis. Switch datasheets state a per-cage power budget for a reason: a line card of QSFP28 ports is a thermal design, not just a port count. When comparing hosts across generations, compare the cage power envelope with the same seriousness as the port rate — a cage that cannot feed the module class you plan is a ceiling no firmware will lift.
There is a second, operational difference: density economics. SFP and SFP+ put one link in a small face; QSFP28 puts four lanes behind one face, and with breakout those lanes can serve four servers. Face-plate density is why the QSFP form factor owns the spine even where per-lane rates equal SFP28’s: the front of the switch is the scarcest real estate in the building.
Which form factor should a new access-layer design choose?
For access and management, SFP and SFP+ remain the economical standards. Choose the cage your rate plan needs for the host’s service life — and no higher; unused rate headroom in a cage is money parked in the faceplate.
Do DAC assemblies follow the same families?
Yes. DAC and AOC assemblies are terminated in the same form factors — an SFP+ DAC and a QSFP28 DAC are different animals despite the shared idea. Order assemblies by the cage they must plug into.
A buyer’s field guide to cages
Stand in front of an unfamiliar switch and you can read its generation from its face. A wall of small square cages with duplex LC pairs is the SFP/SFP+ world — access, aggregation, server-facing ports. Wider cages with MPO or duplex LC and visibly heavier latching are QSFP territory — spine and uplink country. The reading matters because optics orders go wrong at exactly this glance: a purchase list written from a spreadsheet ("the 10G switches") meets a face full of QSFP28 ports installed in last year’s refresh. The professional habit is photograph-the-faceplate: before any optics order, one photograph of each host type, attached to the purchase conversation. It takes a minute and has prevented more wrong-module shipments than any datasheet cross-check ever written.
And a last naming trap: marketing has reused these letters loosely for years ("SFP28" ports described as "SFP ports", QSFP+ (40G) confused with QSFP28 (100G)). The letters after Q and the number after the family are the specification. Read them literally, order them literally, and let no one abbreviate them away in a purchase order.