QSFP (Quad Small Form-factor Pluggable) is the dominant high-density optical module specification in modern data centers, where the “Quad” prefix indicates four electrical lanes per module, delivering four times the per-port bandwidth of a single-lane SFP. From 40G QSFP+ over 4×10G NRZ lanes, through 100G QSFP28 over 4×25G, to 400G QSFP-DD and QSFP112 driven by 8×50G or 4×100G PAM4, the QSFP family has become the backbone interface of leaf-spine fabrics, AI training clusters, and hyperscale interconnects. This in-depth guide walks through the rates, form factors, MSA standards, optical reach, power budgets, breakout architectures, and selection logic of the QSFP+ / QSFP28 / QSFP-DD / QSFP112 family — written from the perspective of engineers who code, test, and ship these modules every day at Sanoc’s own factory in Taiwan.
QSFP Family: 40G / 100G / 400G
| Specification | Rate | Channels | Typical Use | Sanoc Products |
|---|---|---|---|---|
| QSFP+ | 40G | 4×10G | spine interconnect, HPC | 40G QSFP+ |
| QSFP28 | 100G | 4×25G | Data center backbone, cloud | 100G QSFP28 |
| QSFP-DD / QSFP112 | 400G | 8×50G / 4×100G | Hyperscale, AI training backbone | Inquire about 400G/800G series |
What is QSFP28? = A 100G specification QSFP that contains 4×25G channels, currently the most popular 100G interface in data centers. QSFP-DD (Double Density) = Doubles the channels on the QSFP28 form factor to achieve 400G, while being backward compatible with QSFP28.
What “Quad” Actually Means in the Electrical Lane Structure
Every QSFP variant exposes four (QSFP+/QSFP28) or eight (QSFP-DD) high-speed electrical lanes on its edge connector, defined by the relevant MSA and SFF specifications. In QSFP+ each of the four lanes runs at 10.3125 Gbit/s (10GBASE-R), aggregating to 40G. QSFP28 raises each lane to 25.78125 Gbit/s (25G NRZ per IEEE 802.3by), aggregating to 100G. QSFP-DD adds a second row of contacts to carry eight lanes — at 50G PAM4 (per IEEE 802.3cd / 802.3bs) that is 8×50G = 400G, and at 100G PAM4 (IEEE 802.3ck) it becomes 4×100G or 8×100G for 400G/800G classes. The lane count and per-lane modulation are the single most important attributes when you reason about breakout, FEC, and host channel compliance.
40G vs 100G vs 200G vs 400G at a Glance
The generational jump is not just “more bandwidth.” It is also a shift in modulation (NRZ → PAM4), in forward error correction (no FEC / KR4-FEC → RS-FEC), and in connector (MPO-12 → MPO-12/16, Duplex LC → dual-CS or SN). 40G and early 100G are NRZ with relatively relaxed link budgets; 200G/400G are PAM4 and mandate RS(544,514) RS-FEC to recover the higher symbol-error rate inherent to four-level signalling.
QSFP vs QSFP28 vs QSFP-DD vs QSFP56 Form-Factor Map
| Form factor | Lanes × rate | Aggregate | Modulation | Governing MSA / IEEE |
|---|---|---|---|---|
| QSFP+ | 4×10G | 40G | NRZ | SFF-8436, IEEE 802.3ba |
| QSFP28 | 4×25G | 100G | NRZ | SFF-8665, IEEE 802.3bm/by |
| QSFP56 | 4×50G | 200G | PAM4 | SFF-8665, IEEE 802.3cd |
| QSFP-DD | 8×50G or 4×100G | 400G / 800G | PAM4 | QSFP-DD MSA, IEEE 802.3bs/cd/ck |
| QSFP112 | 4×100G | 400G | PAM4 | QSFP-DD/112 MSA, IEEE 802.3ck |
QSFP-DD is physically backward compatible with QSFP28/QSFP56: a QSFP28 module drops into a QSFP-DD cage and uses only the first row of lanes. This compatibility is exactly why hyperscalers standardized 400G ports on QSFP-DD — the same cage can host legacy 100G optics during migration.
Common QSFP28 Models: SR4 / LR4 / ER4 / CWDM4 / PSM4 / DAC / AOC
| Model | Fiber / Interface | Distance | Applicable |
|---|---|---|---|
| 100G SR4 | Multi-mode MPO-12 | ~100m | In-rack / Short-range same row |
| 100G LR4 | Single-mode Duplex LC | 10km | Cross-building / Campus |
| 100G ER4 | Single-mode LC | 40km | Metro long-distance |
| 100G CWDM4 / PSM4 | Single-mode | 2km | Medium distance, cost optimized |
| 100G DAC / AOC | Copper cable / Active optical cable | ≤7m / ≤30m | In-rack break-out |
Common deployment: Using 100G QSFP28 for spine, paired with 25G SFP28 for leaf (100G can break out into 4×25G), is the standard configuration for modern leaf-spine architectures. Short-range interconnects can use DAC copper cables or AOC optical cables to further reduce costs.
Quantified Bench Data: Wavelength, Reach, Power Budget, and Power Draw
The table below summarizes typical optical parameters our test team measures on Sanoc QSFP variants during pre-shipment validation. “Spec reach” is the IEEE/MSA worst-case; “Measured reach” is the typical error-free distance we observe on clean OS2/OM4 plant at room temperature with RS-FEC enabled where applicable. These are representative figures for engineering planning — always confirm against the specific datasheet for a part number.
| Variant | Wavelength | Spec reach | Measured reach (typ.) | Link budget (dB) | Power draw (typ.) | Lanes / connector |
|---|---|---|---|---|---|---|
| 40G QSFP+ SR4 | 850 nm | 150 m (OM4) | 120–150 m | ~1.9 | ~1.5 W | 4×10G / MPO-12 |
| 40G QSFP+ LR4 | 1271–1331 nm CWDM | 10 km | 10–12 km | ~6.7 | ~3.5 W | 4×10G / Duplex LC |
| 100G QSFP28 SR4 | 850 nm | 100 m (OM4) | 70–100 m | ~1.9 | ~2.5 W | 4×25G / MPO-12 |
| 100G QSFP28 PSM4 | 1310 nm | 500 m–2 km | 500 m–2 km | ~3.3 | ~3.5 W | 4×25G / MPO-12 |
| 100G QSFP28 CWDM4 | 1271–1331 nm | 2 km | 2 km | ~5.0 | ~3.5 W | 4×25G / Duplex LC |
| 100G QSFP28 LR4 | 1295–1310 nm LAN-WDM | 10 km | 10–12 km | ~8.5 | ~3.5–4.5 W | 4×25G / Duplex LC |
| 100G QSFP28 ER4 | 1295–1310 nm | 40 km | 30–40 km | ~18 | ~4.5–5.0 W | 4×25G / Duplex LC |
| 400G QSFP-DD DR4 | 1310 nm | 500 m | 500 m | ~3.0 | ~10–12 W | 8×50G PAM4 / MPO-12 |
| 400G QSFP-DD FR4 | 1271–1331 nm | 2 km | 2 km | ~4.0 | ~10–12 W | 8×50G PAM4 / Duplex LC |
| 400G QSFP-DD LR4 / 2×100G | 1271–1331 nm | 10 km | 10 km | ~6.3 | ~12 W | 8×50G PAM4 / Duplex LC |
Reading the Power Budget: Why dBm Margin Matters
Link budget is the difference between minimum launch power and receiver sensitivity. A 100G LR4 with an ~8.5 dB budget over a 10 km span that exhibits roughly 0.3 dB/km fiber loss plus 1–2 dB of connector/splice loss leaves a comfortable few dB of margin. When you breakout, splice, or extend beyond spec, run the numbers in dBm first. Our companion mW to dBm conversion tool and the optical power budget methodology in our knowledge center help you verify margin before you order optics rather than after a link flaps in production.
MPO-12, MPO-16, and Duplex LC — Picking the Right Connector
Parallel-optic variants (SR4, PSM4, DR4) use MPO-12 ribbon with eight active fibers (4 Tx + 4 Rx). 400G 8-lane parallel optics such as DR8 use MPO-16. WDM variants (CWDM4, LR4, FR4) multiplex all lanes onto a single fiber pair and use Duplex LC. The practical takeaway: parallel optics make per-lane breakout trivial (each fiber pair maps to one downstream port), while WDM optics give you the cleanest two-fiber cabling but cannot be split optically.
Breakout Architecture: QSFP → 4×SFP
Breakout is one of the most cost-effective levers in a leaf-spine fabric. A single 100G QSFP28 port can be configured as 4×25G, and a 40G QSFP+ port as 4×10G, quadrupling effective port count without buying more switch ASIC ports.
Breakout Cabling Options and First-Hand Wiring Notes
There are three common ways to break out: (1) a QSFP28-to-4×SFP28 DAC breakout cable for sub-7 m in-rack runs, (2) a QSFP28-to-4×SFP28 AOC breakout for sub-30 m intra-row, or (3) a parallel SR4/PSM4 transceiver fanned out with an MPO-to-4×LC harness to four discrete SFP28 modules. From the field: always confirm the switch supports the chosen breakout mode in software (e.g., interface breakout module 1/1 map 4x25G on Cisco NX-OS, or the equivalent profile on Arista EOS) before cabling — the physical fanout only works if the ASIC port is set to 4×25G mode, otherwise the port stays administratively in 1×100G and the sub-ports never appear.
When Breakout Helps and When It Hurts
Breakout shines for top-of-rack 25G server aggregation under a 100G spine, and for connecting older 10G server pools beneath a 40G distribution layer. It hurts when you need full 100G to a single endpoint, or when FEC/auto-negotiation mismatches between the QSFP host and the SFP28 endpoints cause link-up failures — a frequent gotcha where the 25G endpoint defaults to RS-FEC but the breakout sub-port does not, or vice versa.
Modulation, FEC, and Signal Integrity
PAM4 vs NRZ — the 200G/400G Inflection
NRZ (non-return-to-zero) encodes one bit per symbol with two voltage levels; PAM4 encodes two bits per symbol with four levels, doubling throughput at the same baud rate but cutting the eye-opening into thirds and roughly tripling susceptibility to noise. That is why 40G and most 100G optics are NRZ and FEC-optional, while every 50G-per-lane and 100G-per-lane class (200G QSFP56, 400G QSFP-DD, QSFP112) is PAM4 and FEC-mandatory.
RS-FEC: Mandatory at 50G+ per Lane
Reed-Solomon RS(544,514) FEC — commonly called KP4 FEC — is required by IEEE 802.3cd/bs/ck for PAM4 links. It corrects the higher raw bit-error rate of PAM4 down to an operational post-FEC BER below 1e-12. The practical engineering consequence: end-to-end FEC mode must match across the QSFP, the DAC/AOC, and the far-end host. A 400G link with FEC enabled on one side only will either fail to come up or take continuous corrected/uncorrected errors visible in DDM counters.
CMIS and SFF Management Interfaces
Lower-speed QSFP modules expose their EEPROM and DDM telemetry through the SFF-8636 management memory map (QSFP+/QSFP28), while 400G QSFP-DD and modern modules use CMIS (Common Management Interface Specification). CMIS adds a state machine for application selection, datapath control, and richer telemetry — relevant when a switch advertises a module as “unsupported” simply because it expects CMIS but the optic presents legacy SFF-8636, or the reverse. The relevant SFF documents engineers reference in practice include SFF-8636 for the QSFP+/QSFP28 management map, SFF-8665 for the QSFP28 hardware specification, and SFF-8661 for the QSFP+ mechanical envelope; CMIS revisions 4.x and 5.x govern the 400G/800G generation. When you debug a “module inserted but datapath down” condition on a QSFP-DD link, the CMIS state machine is the first place to look — the host must explicitly provision the datapath, unlike the simpler plug-and-go behaviour of legacy SFF-8636 optics.
Host Channel Compliance and Pre-Emphasis
At 25G NRZ and especially at 50G/100G PAM4, the electrical channel between the switch ASIC SerDes and the module connector matters as much as the optics. IEEE defines host and module compliance test points so that a compliant module works on any compliant host. In the field, marginal links at 100G+ are frequently fixed not by swapping optics but by adjusting the host SerDes pre-emphasis/equalization or by shortening the PCB trace budget with a better-quality cage. This is why Sanoc validates each module against real switch SerDes rather than only against a BERT — a module can pass a clean lab channel yet struggle on a long host trace, and vice versa.
Key Points for QSFP Selection
- Rate: 40G→QSFP+, 100G→QSFP28, 400G→QSFP-DD
- Distance: <100m Multi-mode SR4 / 10km Single-mode LR4 / 40km ER4
- Interface: Multi-mode uses MPO-12, Single-mode uses Duplex LC
- Break-out: 100G QSFP28 can break out to 4×25G, 40G QSFP+ can break out to 4×10G, increasing port density
Further reading: Compatible Optical Module Selection Guide, Arista Compatible QSFP28 Guide.
A Practical QSFP Selection Decision Tree
Step 1 — fix the rate from the switch port (40G/100G/200G/400G). Step 2 — measure the link distance and choose reach class: under 100 m intra-row prefer DAC/AOC or SR4; up to 2 km prefer PSM4/CWDM4/FR4; up to 10 km LR4; up to 40 km ER4. Step 3 — match the connector and fiber plant you already own (MPO-12 ribbon vs Duplex LC vs single-fiber pair). Step 4 — confirm modulation and FEC compatibility with both hosts (NRZ no-FEC vs PAM4 RS-FEC). Step 5 — decide native vs breakout based on the endpoint speed. Step 6 — confirm vendor compatibility coding for the host brand. Following these six steps in order eliminates the vast majority of “module not recognized” and “link won’t come up” tickets.
Identifying a QSFP and Reading DDM in the CLI
To confirm what is plugged in and how healthy it is, read the transceiver inventory and digital diagnostics directly from the switch. On Cisco: show interface Ethernet1/1 transceiver details exposes the QSFP type, vendor name, part/serial number, and per-lane Tx/Rx power, bias current, temperature, and supply voltage. On Arista EOS the equivalent is show interfaces Ethernet1/1 transceiver. From experience, the first things we check on a flapping link are: (a) per-lane Rx power within the receiver sensitivity window, (b) temperature under the module’s high-alarm threshold, and (c) that all four (or eight) lanes show power — a single dark lane usually means a dirty MPO fiber or a bent ribbon.
Thermal and Power Considerations in Dense Deployments
Module power scales sharply with speed: a 100G QSFP28 LR4 draws ~3.5–4.5 W, but a 400G QSFP-DD optic draws ~10–12 W — and a fully populated 32-port 400G line card therefore dissipates several hundred watts of optics alone. In high-density racks we have seen modules hit their high-temperature alarm not because of a defect but because of upstream airflow blockage or a reversed airflow (port-side intake vs exhaust) switch facing the wrong way. Practical mitigations: confirm the switch airflow direction matches the cold/hot aisle, keep blanking panels installed, and treat the module case-temperature DDM reading as an early-warning sensor for the whole rack, not just the optic.
Power Class and Cage Cooling
QSFP-DD and CMIS define module power classes (Class 1 through Class 8), and a switch port advertises the maximum power class it can cool. If you insert a high-power-class 400G ZR/ZR+ coherent optic into a port provisioned for a lower class, the host may keep the module in low-power mode or refuse to bring up the datapath. When planning a 400G refresh, confirm both the per-port and the per-line-card power budget — not just the optical reach — because a switch can be optically capable yet thermally constrained from running every port at full power simultaneously.
DDM Thresholds and Proactive Monitoring
Every QSFP exposes warning and alarm thresholds for temperature, supply voltage, Tx bias, Tx power, and Rx power in its diagnostic memory. The disciplined operational practice is to poll these DDM values via SNMP, gNMI, or streaming telemetry and alert on the warning threshold rather than waiting for the alarm. From experience, a slow drift of Rx power toward the low-warning limit over weeks almost always indicates a dirty or degrading connector that, left alone, will eventually cross the low-alarm threshold and drop the link during peak traffic.
Is Compatible QSFP Safe?
Safe. QSFP follows MSA, and Sanoc compatible QSFP+ / QSFP28, after Sanoc FreeCode free EEPROM encoding, functions identically to original manufacturers on devices such as Cisco, Arista, Juniper, HPE, etc., significantly reducing costs without affecting original warranties, making it particularly suitable for large data center deployments.
How EEPROM Compatibility Coding Works
A “compatible” QSFP is an MSA-standard module whose SFF-8636/CMIS EEPROM has been programmed with the vendor ID, part number, and compliance fields that a given switch OS expects to see. Because the optical and electrical behaviour is defined by the MSA and IEEE standards — not by the brand — a correctly coded compatible module is electrically and optically equivalent to the OEM part. Sanoc FreeCode programs this at no extra cost and validates it on real Cisco/Arista/Juniper/HPE hardware before shipment.
Why Choose Sanoc’s QSFP Modules
- Complete coverage of 40G QSFP+ / 100G QSFP28 / 400G
- Cisco/Arista/Juniper/HPE compatibility guarantee + free EEPROM encoding + pre-shipment real machine testing
- 3-year warranty · DOA immediate replacement · Does not affect original warranty · Bulk discounts
- Own factory in Taiwan · 4-hour quotation
Frequently Asked Questions
What is the difference between QSFP+ and QSFP28?
QSFP+ runs four 10G NRZ lanes for a 40G aggregate, while QSFP28 runs four 25G NRZ lanes for a 100G aggregate. They share the same physical form factor and cage, but a 100G QSFP28 will not negotiate 100G in a 40G-only host port. Always match the port’s supported speed.
Can a QSFP28 module fit in a QSFP-DD port?
Yes. QSFP-DD cages are backward compatible with QSFP28 and QSFP56. A QSFP28 module seats into a QSFP-DD port and uses only the first row of electrical lanes, which is why QSFP-DD is the preferred 400G form factor for smooth migration from existing 100G optics.
How far can a 100G QSFP28 reach?
It depends on the variant: SR4 multimode reaches ~100 m on OM4, PSM4/CWDM4 reach 500 m–2 km on single-mode, LR4 reaches 10 km, and ER4 reaches up to 40 km. DAC copper handles ≤7 m and AOC handles ≤30 m for in-rack interconnect.
What connector does each QSFP variant use?
Parallel optics (SR4, PSM4, DR4) use MPO-12 ribbon (DR8 uses MPO-16). WDM optics (CWDM4, LR4, FR4, ER4) multiplex all lanes onto a single fiber pair and use Duplex LC. Choose based on the fiber plant you already have.
Do I need RS-FEC for 400G QSFP-DD?
Yes. All 50G-per-lane and 100G-per-lane PAM4 classes (200G QSFP56, 400G QSFP-DD, QSFP112) mandate RS(544,514) RS-FEC per IEEE 802.3cd/bs/ck. FEC mode must match end to end across the optic, the DAC/AOC, and the far-end host, or the link will error or fail to come up.
What is the difference between PSM4 and CWDM4 at 100G?
Both reach roughly 500 m–2 km on single-mode fiber. PSM4 uses four parallel single-mode fibers (MPO-12) at one wavelength and is breakout-friendly; CWDM4 multiplexes four wavelengths onto one Duplex LC fiber pair, saving fiber count but not splittable. PSM4 is cheaper on optics but needs more fiber; CWDM4 is the opposite trade-off.
Can I break out a 100G QSFP28 into 4×25G SFP28?
Yes — using a QSFP28-to-4×SFP28 DAC/AOC breakout cable or an MPO-to-4×LC harness with parallel optics. You must first set the switch ASIC port to 4×25G breakout mode in software; the physical fanout alone will not bring the sub-ports up. Also confirm FEC settings match between the breakout sub-ports and the 25G endpoints.
How much power does a QSFP module consume?
Roughly 1.5 W for 40G SR4, 2.5–4.5 W for most 100G QSFP28 variants, and 10–12 W for 400G QSFP-DD. In dense racks the aggregate optics power per line card can reach several hundred watts, so airflow direction and thermal headroom matter as much as the optic itself.
How do I read a QSFP’s diagnostics from the switch?
Use the transceiver detail command — for example show interface Ethernet1/1 transceiver details on Cisco NX-OS or show interfaces Ethernet1/1 transceiver on Arista EOS. This shows vendor, part/serial number, temperature, supply voltage, and per-lane Tx/Rx power and bias current, which are the first values to check when troubleshooting a flapping link.
Is a compatible (third-party) QSFP safe to use with Cisco or Arista switches?
Yes. QSFP optics follow the MSA and IEEE standards, so a correctly EEPROM-coded compatible module behaves identically to the OEM part. Sanoc FreeCode programs the vendor compatibility fields for free and validates each module on real Cisco/Arista/Juniper/HPE hardware before shipment, reducing cost without affecting your switch warranty.
What is QSFP112 and how is it different from QSFP-DD?
QSFP112 uses four 100G PAM4 lanes (4×100G = 400G) on a single-row QSFP form factor, whereas QSFP-DD uses eight 50G PAM4 lanes (8×50G = 400G) on a double-density two-row connector. QSFP112 trades lane count for higher per-lane rate and is the building block for next-generation 800G modules built from two 400G electrical groups.
Data Center 40G/100G/400G Deployment?
Free assistance in selection + free test samples + bulk discount quotations.
Related In-Depth Guides
Author: Sanoc Optical Communications Engineering Team — SANway Optoelectronics (Sanoc) is a Taiwan-based B2B optical transceiver manufacturer with its own factory in Hsinchu, specializing in compatible SFP / SFP+ / SFP28 / QSFP / QSFP28 modules for Cisco, Arista, Juniper, HPE, MikroTik and other major platforms. Winner of the 2025 Taiwan Excellence Award.
Technical basis: This article follows the MSA (Multi-Source Agreement), IEEE 802.3 Ethernet standards and ITU-T optical recommendations.
Quality & review: All Sanoc modules are bench-tested on enterprise-grade switches before shipping, with a 1-year warranty and immediate DOA replacement, without voiding your switch warranty. Contact our engineers with any questions.
Last updated: June 2026 | Educational content; engineering inquiries are replied to within 4 hours.
Further Reading: Expert Technical Columns
- Cisco Compatible SFP & SFP+: The Complete Compatibility Guide
- Do Compatible Transceivers Void Your Warranty? The Engineering Answer
- Arista, Juniper and HPE Aruba Compatible Transceivers: Platform Notes
- IEEE 802.3 and the MSA: What Transceiver Standards Actually Guarantee
- The 400G to 800G Data Center Transition: What IT Leaders Should Plan For
- AI Networking and the Optical Interconnect Surge: A Strategic View
- My SFP Link Won’t Come Up — A Field Troubleshooting Guide
- Inside the Sanoc QA Lab: How We Bench-Test Every Batch
- Why Taiwan Optical Manufacturing Matters for Your Supply Chain
