QSFP28 vs QSFP-DD: A Complete Technical Guide for 2026

Look, if you're spec'ing out a data center refresh or planning your next spine-leaf buildout, you've probably hit this exact question: stick with tried-and-true 100G QSFP28, or make the jump to 400G QSFP-DD?

It's not just about bandwidth numbers on a spec sheet. The form factor you choose today locks you into a specific power envelope, cooling strategy, and cabling infrastructure for the next 5-7 years. Get it wrong, and you're either sitting on stranded capacity or bleeding money on tech your traffic doesn't need yet.

Let's break this down the way we'd talk through it over coffee.

 

 

The Physical Guts: What Actually Changed

Here's the deal with QSFP-DD-the "DD" stands for Double Density, and that's exactly what they did. They took the original QSFP28 connector and slapped a second row of electrical contacts behind it.

QSFP28 gives you four lanes running at 25 Gbps each (NRZ modulation), totaling 100G per port. Same width it's always been: 18.35mm. QSFP-DD doubles those lanes to eight, each pushing 50G with PAM4 signaling, hitting 400G aggregate.

The clever bit? That second connector row adds only about 2.5mm to the module depth. Width stays the same, so your existing port spacing still works.

Here's the compatibility kicker: A QSFP-DD cage happily accepts your old QSFP28 modules-they just use the front row of contacts. But flip it around? No dice. QSFP-DD modules physically won't seat in QSFP28-only ports because of that extended depth.

Spec	QSFP28	QSFP-DD
Electrical Lanes	4	8
Max Lane Rate	25G NRZ / 50G PAM4	50G PAM4 / 100G PAM4
Aggregate Bandwidth	100G / 200G	400G / 800G
Module Width	18.35mm	18.35mm
Module Depth	~70mm	~87mm
Connector Rows	Single	Dual

This backward compatibility is huge for phased migrations-more on that later when we talk ROI.

 

 

The Heat Problem Nobody Wants to Talk About

This is where things get real.

A typical QSFP28 module at 100G pulls between 2.5W and 3.5W. Standard passive heatsinks with decent front-to-back airflow? You're golden. Most switching platforms handle this without breaking a sweat.

QSFP-DD at 400G? We're talking 10W to 15W per module. Extended reach variants can push 18W.

Let that sink in for a second. That's roughly a 4x increase in thermal dissipation. Per port.

To put it in real-world terms: imagine your current server room AC is sized to cool a sedan's worth of heat per rack. Now you're asking it to handle a pickup truck. Same footprint, 4x the thermal load.

What this means for your infrastructure:

The airflow math changes completely. QSFP28 ports typically need 20-25 CFM (cubic feet per minute) per position. QSFP-DD? You're looking at 40-60 CFM. That means faster fans, more fans, or both.

For a 48-port switch, the numbers look like this:

Configuration	Power Draw (Transceivers Only)
48x QSFP28 @ 3.5W	168W
48x QSFP-DD @ 15W	720W

That 552W delta per switch adds up fast across a pod. If you're running conventional raised-floor cooling, you might be looking at rear-door heat exchangers or enhanced cold aisle containment to make 400G work.

 

 

PAM4: The Magic (and Complexity) Behind 400G

Okay, quick detour into the signal layer-because this explains a lot about why 400G costs more and runs hotter.

QSFP28 uses NRZ signaling. Simple stuff: voltage high = 1, voltage low = 0. One bit per symbol. At 25 Gbps per lane, it's clean and doesn't need much signal conditioning.

QSFP-DD uses PAM4. Instead of two voltage levels, you get four. Each symbol encodes two bits. You've effectively doubled your bit rate without cranking up the symbol rate-which matters because higher symbol rates hit nasty signal integrity walls from trace losses and crosstalk.

Think of it like this: NRZ is a light switch-on or off. PAM4 is a dimmer switch with four positions. You're packing more information into each "click," but the receiver has to be way more precise about detecting which position you're at.

The trade-off? Those four voltage levels are squeezed into the same overall signal swing. Each "step" between levels is only about 1/3 the height of NRZ. That's roughly 9.5 dB less noise margin-which is why FEC (Forward Error Correction) isn't optional at 400G, it's mandatory.

Parameter	NRZ (QSFP28)	PAM4 (QSFP-DD)
Bits per Symbol	1	2
Symbol Rate for 50G	50 GBaud	25 GBaud
Eye Amplitude	Full swing	~1/3 of full swing
SNR Requirement	Lower	~10 dB higher
FEC	Optional	Required (KP4)

The KP4 FEC (RS-544,514) adds about 2.6% bandwidth overhead and 50-100ns latency per hop. For most applications, you won't notice. For latency-sensitive trading workloads? That's a conversation worth having.

All this PAM4 signal processing-equalization, FEC encoding/decoding, clock recovery across eight lanes-requires dedicated DSPs burning 4-6W just doing math. That's a big chunk of why 400G modules run so much hotter than their 100G predecessors.

 

 

The Backward Compatibility Story

Here's where QSFP-DD earns its keep for anyone doing a phased migration.

That dual-row connector design we talked about? It means your QSFP-DD switches can run your existing QSFP28 optics at full 100G speeds. The host detects the module type during init and configures the port accordingly. No adapter dongles, no weird compatibility modes.

"QSFP-DD's backward compatibility with QSFP28 allows network operators to leverage existing 100G optics inventory while deploying new 400G switching platforms, significantly reducing migration costs." - Ethernet Alliance, QSFP-DD Technology Overview

Practical scenarios this enables:

Your new 400G spine switch can connect to legacy 100G leaf switches AND new 400G leaves simultaneously. Same chassis, mixed speeds, zero drama.

Or: you deploy QSFP-DD platforms now, populate them with your existing QSFP28 inventory, and upgrade individual links to 400G as traffic demands it. No forklift upgrade required.

If you've got 500 QSFP28 modules sitting in inventory-representing maybe $250K-$500K in hardware-that's not stranded investment. Those optics keep working in your new infrastructure.

A few gotchas to keep in mind:

QSFP28 modules max out at 100G (or 200G with newer 50G PAM4 variants) regardless of what the host can do

Thermal systems optimized for 400G might over-cool QSFP28 modules-in humid environments, that can cause condensation. Some platforms let you dial back per-port fan speeds.

Fiber connector types may differ between your 100G and 400G implementations-verify separately

For teams managing mixed 100G/400G environments, we offer QSFP28 modules and QSFP-DD transceivers with consistent quality standards across both form factors-simplifies your qualification process during transition.

 

 

Cabling: New Connectors Enter the Chat

When you jump from 100G to 400G, your fiber connectivity options shift too.

What you're probably running now (QSFP28):

LC Duplex for single-fiber-pair applications (100G-LR4, 100G-CWDM4). WDM crams four 25G channels onto one fiber pair-standard LC patch panels handle it fine.

MPO-12 for parallel stuff (PSM4, SR4). Four fiber pairs for TX/RX, with four fibers in that 12-ribbon sitting unused. Not exactly efficient, but it works.

What 400G brings to the table:

MPO-16 for parallel 400G (SR8 and friends). Eight TX, eight RX, zero unused fibers. Finally, full ribbon utilization.

MPO-12 with breakout still works for some 400G applications. Modules like 400G-DR4 use four 100G-per-lambda channels, matching the MPO-12 physical layer but quadrupling aggregate bandwidth.

CS (Compact Simplex) and SN connectors are the new hotness for high-density single-fiber runs. About half the panel space of LC-serious fiber count improvements per RU.

If you're speccing new structured cabling, think ahead. Your fiber plant should accommodate the full range of connector types you'll realistically need over the next 5-7 years.

The DAC reach problem at 400G:

Direct Attach Cables are great for short-haul, same-row connections. But the signal integrity challenges at PAM4 rates cut your reach significantly.

Cable Type	QSFP28 100G	QSFP-DD 400G
Passive DAC	Up to 5m	Up to 2.5m
Active DAC	Up to 7m	Up to 3m
AOC	Up to 100m	Up to 100m

That 2.5m passive DAC limit at 400G might push more of your connections toward AOCs or pluggable transceivers. Factor that into your per-port cost modeling.

We stock DAC cables and AOC options across both QSFP28 and QSFP-DD form factors if you're standardizing your interconnect strategy.

 

 

So When Do You Actually Make the Jump?

After all the specs and thermal math, here's how we'd frame the decision.

Stick with QSFP28 if:

Your 100G links are running below 50% sustained utilization. You've got headroom.

Power and cooling constraints box you in, and a 4x thermal increase would mean facility upgrades you're not ready for.

Your QSFP28 inventory is substantial and still operationally valuable.

You're talking edge or branch locations where traffic aggregates elsewhere anyway.

Move to QSFP-DD if:

Rack space is at a premium and you need maximum port density (4x bandwidth, same front-panel footprint).

Your spine-leaf architecture needs higher oversubscription ratios to handle dense leaf tiers.

You're building a new facility where you can right-size power and cooling from day one.

Competitive or stakeholder pressure requires demonstrating next-gen capabilities.

The hybrid path most orgs actually take:

Deploy QSFP-DD switching platforms with backward-compatible ports. Populate with a mix of QSFP28 and QSFP-DD transceivers based on actual link requirements. Upgrade individual connections as traffic grows.

This gives you the flexibility to source optics at competitive prices while preserving your upgrade path. It's not either/or-it's "both, strategically."

Deployment Type	Recommendation	Why
Edge/Branch	Stay QSFP28	Lower power, bandwidth is plenty
Enterprise DC	Phased migration	Protect investment, grow gradually
Cloud/Colo	Accelerate to QSFP-DD	Density pressure, competitive positioning
Hyperscale	Full QSFP-DD	Port density, operational consistency

 

 

 

What's Coming: 800G on the Same Platform

One more reason QSFP-DD makes sense for forward-looking deployments: the form factor roadmap extends to 800G.

By pushing 100G-per-lane PAM4 across eight lanes, early 800G QSFP-DD modules hit the market in 2024. Broader adoption is expected through 2026-2027 as switch ASIC support matures.

OSFP is the competing 800G form factor-slightly larger with higher power capacity. The industry hasn't reached consensus on long-term direction yet, but QSFP-DD's backward compatibility gives it a real edge for enterprise adoption.

Bottom line: if you buy QSFP-DD switching platforms today, you're positioned for 800G upgrades through transceiver swaps alone. Your switch hardware investment carries across multiple bandwidth generations.

 

 

Wrapping Up

The QSFP28 vs QSFP-DD decision isn't really about 100G vs 400G bandwidth. It's about matching your infrastructure investment to your actual requirements-current and projected.

QSFP28 still makes sense where 100G is plenty and you want to sweat existing assets. QSFP-DD delivers the density hyperscale environments demand while giving enterprise shops a migration path that doesn't strand their 100G inventory.

Most successful deployments we see combine strategic platform selection with tactical flexibility in optics procurement. Understand the thermal implications, plan your cabling infrastructure for the connector evolution, and you'll be set up for whatever traffic growth throws at you.

 

 

---

Technical specifications vary by manufacturer and module configuration. Always verify with vendor datasheets before procurement. Market projections represent industry analyst estimates and may not reflect actual conditions.

References:

1. IDC, "Worldwide Enterprise Network Infrastructure Forecast, 2023-2027" (verify current publications for updated projections)
2. QSFP-DD MSA Group, "QSFP-DD Hardware Specification Rev 6.0"
3. Ethernet Alliance, "QSFP-DD Technology Overview," 2023
4. Uptime Institute, "Data Center Capacity Planning Best Practices"