What Is QSFP28? 100G Types & Module Selection

QSFP28 is a compact, hot-pluggable transceiver form factor built for 100G Ethernet. It reaches 100Gbps by running four electrical lanes at 25Gbps each, and it is the most widely deployed 100G module type in data centers, enterprise cores, telecom edge networks, and data center interconnect (DCI) links. The 4×25G lane architecture behind it was standardized by the IEEE 802.3bm amendment, which is why the same generation of optics (SR4, LR4, CWDM4, PSM4) all trace back to one common design baseline.

If you are building a fresh 100G link, migrating up from 40G, wiring leaf-to-spine uplinks, or splitting one 100G switch port into four 25G server connections, QSFP28 is usually the first form factor on the table. The harder question is not what QSFP28 is - it is which QSFP28 module matches your distance, fiber plant, and connector. This guide is the foundational answer: what QSFP28 means, how it compares with neighboring form factors, and how to shortlist the right module type.

What Does QSFP28 Stand For?

QSFP28 stands for Quad Small Form-factor Pluggable 28, and the name maps directly to how the module works:

• Quad - four parallel lanes.
• Small Form-factor Pluggable - a compact, tool-free, removable module.
• 28 - roughly 28 gigabaud of per-lane electrical signaling headroom.

The mechanical envelope, cage, connector, and management interface are defined under the SFF-8665 specification maintained by the SNIA SFF committee. Although the name says "28," each lane carries 25Gbps of Ethernet payload; the extra headroom absorbs line coding and FEC overhead. The practical takeaway: QSFP28 is the default form factor for 100G Ethernet, and four 25G lanes are what make that possible.

QSFP28 Key Specifications at a Glance

Real-world reach and behavior still depend on the module design, switch platform, and fiber grade, so the module datasheet is the document that settles the exact numbers.

QSFP28 vs QSFP+, SFP28, QSFP56, and QSFP-DD

These form factors are easy to confuse because several share a similar outer shape. What separates them is lane speed, total bandwidth, and the Ethernet generation they were built for.

QSFP28 vs QSFP+

QSFP+ carries 40G as 4×10G; QSFP28 carries 100G as 4×25G. The cages look nearly identical, but they are not electrically equal. A QSFP+ module may run at 40G in a QSFP28 port if the switch supports the fallback, but the reverse never works: a QSFP28 module cannot produce a 100G link in a QSFP+ port, because that older port cannot clock 25G per lane.

QSFP28 vs SFP28

SFP28 is a single-lane 25G module. QSFP28 bundles four of those lanes into one 100G port. In shorthand: SFP28 = 25G per port, QSFP28 = 4×25G = 100G per port. That relationship is exactly why 100G-to-4×25G breakout is such a common access-layer design - one QSFP28 leaf port feeds four SFP28 server NICs.

QSFP28 vs QSFP56

QSFP56 uses higher-rate 50G PAM4 lanes to reach 200G and shows up on newer switching platforms. Many QSFP56 ports accept a QSFP28 module at 100G, but that backward operation is platform-specific and should be confirmed rather than assumed from the physical fit.

QSFP28 vs QSFP-DD

QSFP-DD ("double density") adds a second row of contacts for eight lanes, targeting 400G and above in greenfield hyperscale and AI clusters. For a deeper breakdown of that generation, see this QSFP-DD technical overview. For mature 100G aggregation, enterprise cores, and cost-sensitive upgrades, QSFP28 remains the pragmatic choice; QSFP-DD or OSFP make more sense only when the whole design is moving to 400G.

Common 100G QSFP28 Module Types: SR4, PSM4, CWDM4, LR4, ER4

"QSFP28" is the form factor; the optical type inside is what actually decides your fiber, connector, and reach. The single most expensive mistake in module selection is usually not the wrong speed - it is the wrong optical interface for the fiber you already have. Distance and fiber type narrow the field faster than anything else, so start there. The table below is the fastest way to shortlist a module before reading the individual notes.

Reach figures are typical, not guaranteed - OM3 versus OM4 grade and connector loss both shift the real budget. Treat the numbers as a starting shortlist and confirm the exact value on the datasheet.

SR4 and the multimode short-reach case

SR4 is the low-cost workhorse inside a single data hall, running 100G over parallel OM3/OM4 with an MPO/MTP connector. It is often the cheapest module on paper, but the MPO cabling and polarity management add complexity that duplex options avoid - something worth weighing if your plant is mostly LC. If you are unsure whether your existing multimode runs will hold 100G at the distance you need, check the multimode fiber reach limits by OM grade before committing.

PSM4 and CWDM4 for the single-mode middle ground

Both cover the gap between short SR4 links and long LR4 hauls, but they solve it differently. PSM4 uses eight parallel single-mode fibers over MPO, which suits short in-facility runs where parallel SMF is already pulled. CWDM4 multiplexes four wavelengths onto a single duplex LC pair, so it reaches roughly 2 km on just two fibers - usually the easier and cheaper option when you are reusing installed duplex single-mode cabling rather than laying new ribbon.

LR4 and the extended-reach ER4 / ZR4

LR4 is the standard answer for 10 km campus and metro backbones on duplex LC single-mode fiber. Beyond that, ER4 (~30–40 km) and ZR4 (~80 km) enter carrier and long-haul territory, where you can no longer skip an optical power budget check - dispersion, receiver sensitivity, and FEC support all become gating factors, and ZR4 in particular is frequently a vendor-specific rather than IEEE-standard implementation.

BiDi and SWDM4 when you must keep existing multimode

BiDi and SWDM4 exist for one reason: reusing an installed duplex multimode plant without recabling to MPO. If ripping out and re-terminating fiber would be expensive or disruptive, these BiDi transceiver designs let you carry 100G over the LC duplex fiber you already own - provided the switch and module coding are confirmed compatible.

How to Choose the Right QSFP28 Module

Selection is a matching problem, not a speed problem. The module has to line up with the switch port, the fiber, the connector, the distance, and the deployment style all at once. Work through the following in order.

Step 1: Confirm the switch port type

Verify whether the port is QSFP+, QSFP28, QSFP56, or QSFP-DD before anything else. Physical fit is misleading - a QSFP28 module can slide into a lookalike cage and still fail if the port cannot supply 25G-per-lane signaling.

Step 2: Measure the real link distance

Use the actual span, not the maximum reach printed in the module name. A "10 km" LR4 on a 200 m run is wasted budget; an SR4 on a 300 m run simply will not link. Short intra-hall hops lean toward SR4, DAC, or AOC; single-mode spans of a few hundred metres to 2 km point to PSM4 or CWDM4; campus and metro reach calls for LR4, ER4, or ZR4.

Step 3: Match the installed fiber

Multimode plant needs SR4, BiDi, or SWDM4; single-mode plant needs PSM4, CWDM4, LR4, ER4, or ZR4. Mismatched fiber is one of the most common causes of a dead link, so treat the installed fiber type as a hard constraint on the module list rather than something to work around.

Step 4: Match the connector

Parallel optics (SR4, PSM4) use MPO/MTP; WDM optics (CWDM4, LR4, ER4, most BiDi) use duplex LC. Check the patch panels, trunks, and distribution frames you actually have, because a connector mismatch stalls deployment even when the optics are correct. If parallel and duplex terminology is causing confusion, this engineer's guide to MTP vs MPO clears up the cabling side.

Step 5: Decide between optical module, DAC, or AOC

For same-rack or adjacent-device links, a QSFP28 DAC is the cheapest and lowest-power option. For slightly longer short-reach runs, AOC is lighter and easier to route than copper. For structured cabling, patch panels, or any distance beyond a few metres, an optical transceiver is the right call.

Step 6: Confirm switch-level compatibility before you order

Mechanical fit is not the whole story. The same optical type can be accepted on one switch and flagged on another because of vendor EEPROM coding, firmware behavior, FEC defaults, and DOM/DDM support - and third-party optics add a further layer. That qualification is a topic in its own right, so rather than compress it here, work through the full breakdown of coding, FEC, firmware, and per-vendor behavior in the dedicated QSFP28 compatibility guide. At minimum, confirm the vendor coding your platform expects, whether RS-FEC is required, and the commercial or industrial temperature rating before placing the order.

QSFP28 Port Compatibility Basics

At the form-factor level, whether two ends will even attempt a link follows simple, mechanical rules based on lane speed:

These cover the physical and lane-speed layer only. Whether a technically-correct module is actually enabled and monitored depends on more than the port - EEPROM coding, firmware, and FEC decide that, and behavior varies across Cisco, Arista, Juniper, Dell, and NVIDIA platforms. For that deeper layer, and for how to qualify third-party optics before buying at scale, the QSFP28 compatibility guide walks through it end to end.

When Should You Use QSFP28 Breakout?

Breakout applies when one 100G port needs to serve four 25G ports - classically, a leaf switch with QSFP28 uplinks fanning out to servers with SFP28 NICs, so you avoid buying separate 25G access switches. It fits well when servers are on 25G, the switch supports 100G-to-4×25G mode, and you want a staged migration path from 25G access to 100G aggregation.

The catch is that breakout is never free at the port level: every port used for a 4×25G split is a port no longer available as a 100G uplink. That trade-off drives real port-budget and oversubscription decisions, which are worked through in detail - including the common mistake of counting the same port twice - in the 100G QSFP28 spine-leaf design guide. On the physical side, the split itself is carried by MPO/MTP breakout cables or breakout DAC/AOC, and breakout support is always switch- and port-group-specific, so confirm it before ordering.

Is QSFP28 Still Worth Using in 2026?

Yes - whenever the requirement is genuinely 100G. Even as 200G and 400G expand at the spine, most enterprise data centers, telecom edge sites, campus cores, and cloud environments still run large amounts of 100G. QSFP28 stays attractive for practical reasons rather than marketing ones: the ecosystem is mature, the optics are available from many vendors, and 100G switch ports and modules are typically cheaper per port than the newest high-speed form factors precisely because the supply chain and standards have been settled since the 802.3bm era.

QSFP28 is the right choice when you are:

• Upgrading from 10G, 25G, or 40G to 100G
• Building stable 100G switch-to-switch uplinks
• Prioritizing broad vendor availability and mature optics
• Not yet ready for a full 200G/400G refresh
• Feeding 25G servers through 100G-to-4×25G breakout

Lean toward QSFP56, QSFP-DD, or OSFP instead when you are designing a greenfield 200G/400G fabric, building large AI or HPC clusters, or your switch roadmap has already moved past QSFP28 density.

FAQ About QSFP28

What is QSFP28 used for?

It carries 100G Ethernet across data centers, enterprise cores, telecom infrastructure, DCI, and high-speed switch uplinks - and, via breakout, connects 100G switches down to 25G servers.

Is QSFP28 the same as 100G?

QSFP28 is the most common 100G form factor, but not the only one. The key buying point is that "100G" alone does not tell you the fiber, connector, or reach - the module suffix (SR4, LR4, CWDM4) does.

Why does QSFP28 use four lanes instead of one?

When QSFP28 was standardized, 25G-per-lane electrical signaling was the practical, cost-effective way to reach 100G, so four 25G lanes were bundled together. Newer single-lambda 100G optics put the full 100G on one wavelength, but the four-lane 4×25G design remains the mature, broadly supported baseline.

What is the difference between QSFP28 and QSFP+?

QSFP+ is 40G (4×10G); QSFP28 is 100G (4×25G). They look similar but are electrically different, so confirm the port type before ordering.

Can QSFP28 work in a QSFP+ port?

No. A QSFP+ port cannot supply 25G-per-lane signaling, so it cannot form a 100G link with a QSFP28 module.

Can QSFP+ work in a QSFP28 port?

Sometimes, at 40G, if the switch supports the fallback - but this is platform-dependent, so check the compatibility matrix rather than assuming.

What is a QSFP28 breakout cable?

It splits one 100G QSFP28 port into four 25G SFP28 connections, commonly used to link 100G switches to 25G servers. It only works if the switch supports and is configured for 4×25G breakout on that port.

Should I choose DAC, AOC, or an optical QSFP28 module?

DAC for same-rack, low-cost links; AOC for slightly longer short-reach runs that benefit from lighter cabling; optical modules for structured cabling, patch panels, and any real distance.

Final Thoughts

QSFP28 remains one of the most practical form factors in networking because it delivers 100G through a mature, well-supported 4×25G design that spans multimode, single-mode, DAC, and AOC options. The real skill is not defining QSFP28 - it is matching the module to your switch port, fiber, connector, and distance.

Pin down those variables first, and the right SR4, CWDM4, LR4, or breakout choice becomes obvious rather than a guess. Once the module type is settled, the two questions that decide success are switch compatibility and, in a fabric, how the ports are budgeted - both of which are worth working through before the purchase order rather than during the first outage.