QSFP28 vs QSFP56 vs QSFP112: Speed, Lane Architecture and Compatibility

1. QSFP28, QSFP56 and QSFP112: Core Specifications

The generation number indicates the approximate electrical signaling rate per lane. Each of these interfaces uses four electrical lanes, but the signaling technology and resulting aggregate bandwidth differ considerably.

Actual electrical interface rates are slightly higher than the commonly quoted 25G, 50G and 100G lane classes because of encoding and interface specifications. For example, a QSFP112 implementation can use four approximately 106.25–112 Gb/s electrical lanes to support a 400GbE link.

2. QSFP28: Four 25G NRZ Lanes for 100G Connectivity

QSFP28 is widely deployed in 100G Ethernet infrastructure. Its four electrical lanes typically operate at approximately 25 Gb/s each, providing an aggregate 100G-class interface.

NRZ, or Non-Return-to-Zero, uses two signal levels to represent binary data. Each symbol carries one bit, making it a relatively straightforward modulation approach for 25G electrical lanes.

Common QSFP28 optical configurations include:

QSFP28 is also available in single-lambda PAM4 optical implementations. Therefore, NRZ describes the common four-lane 100G electrical architecture, not every possible optical modulation format within the QSFP28 product family.

For existing 100G switching infrastructure, QSFP28 remains useful where port density, established optical standards and a broad range of transmission distances are important.

3. QSFP56: Four 50G PAM4 Lanes for 200G Networks

QSFP56 doubles the nominal lane speed compared with QSFP28 while retaining four electrical channels. It typically uses PAM4 signaling to achieve approximately 50 Gb/s per lane, producing a 200G-class aggregate interface.

PAM4 uses four distinct signal levels. Each symbol represents two bits, allowing more information to be transmitted within the same symbol rate compared with NRZ.

Although PAM4 increases transmission efficiency, its four signal levels have smaller voltage spacing than NRZ at a comparable signal amplitude. This makes the link more sensitive to noise, distortion and signal degradation. FEC and appropriate host signal processing are therefore important parts of high-speed PAM4 system design.

QSFP56 is used in 200G Ethernet and high-performance computing environments. Depending on the optical specification, modules can support multimode or single-mode fiber connections, with connector types and reach varying by model.

For network operators upgrading from 100G to 200G, QSFP56 provides a four-lane approach without moving directly to an eight-lane double-density interface.

4. QSFP112: Four 100G-Class Lanes for 400G Connectivity

QSFP112 advances the QSFP interface to approximately 100G-class electrical signaling per lane. Its four-lane architecture supports 400G Ethernet connectivity and is designed for high-bandwidth switching platforms, AI infrastructure and data center interconnection.

Unlike QSFP28's typical 25G NRZ electrical interface, QSFP112 uses high-speed PAM4 signaling. Each lane carries approximately twice the nominal signaling rate of a QSFP56 lane.

Common 400G QSFP112 optical configurations include:

These optical variants use different optical lane arrangements and wavelength plans. The four electrical host lanes should not be confused with the number of optical fibers or wavelengths used by a particular module.

For example, parallel DR4 and SR4 designs use multiple optical channels, while FR4 and LR4 use wavelength-division multiplexing over duplex single-mode fiber.

5. NRZ vs PAM4: The Signaling Difference Behind Each Generation

The transition from QSFP28 to QSFP56 and QSFP112 is closely connected to the evolution of electrical signaling. Increasing the number of bits transmitted per symbol allows higher data rates without requiring a proportional increase in symbol frequency.

NRZ remains suitable for many 100G applications, while PAM4 enables higher data rates within the electrical channel limitations of modern switches and transceivers.

However, a higher modulation level does not automatically guarantee a longer transmission distance. Optical reach still depends on transmitter power, receiver sensitivity, fiber attenuation, dispersion, connector loss and the specific optical standard.

6. QSFP112 vs QSFP-DD: Two Different 400G Architectures

QSFP112 and QSFP-DD can both support 400G applications, but they achieve this bandwidth through different electrical lane configurations.

The key distinction is lane density. QSFP112 uses four faster electrical lanes, whereas the classic 400G QSFP-DD architecture uses eight 50G PAM4 lanes.

This difference affects host SerDes design, connector pin allocation, signal integrity, module implementation and platform qualification. A switch designed for QSFP-DD should not automatically be assumed to support QSFP112 modules, or vice versa.

7. Backward Compatibility and Host Port Requirements

QSFP28, QSFP56 and QSFP112 share the general QSFP mechanical interface concept, but their electrical requirements are not identical. Physical insertion compatibility does not guarantee that a module will establish a working link.

Some QSFP112 host ports are designed to support earlier QSFP generations. For example, certain platforms can operate QSFP112, QSFP56 and QSFP28 modules in the same cage, provided that the hardware, firmware and port configuration support the required operating modes.

Before selecting a module, engineers should verify:

• Switch or network adapter model and supported port generation.

• Supported electrical lane rates and modulation modes.

• Module coding, management interface and firmware requirements.

• Supported optical standards, FEC settings and link configuration.

• Power consumption, thermal limits and airflow requirements.

• Fiber type, connector type and required transmission distance.

For multi-vendor networks, compatibility testing is particularly important. A module that fits a QSFP cage may still be rejected or fail to establish a link if its electrical interface, firmware identification or optical specifications are unsupported.

8. C-LIGHT Optical Transceiver Options for Different Network Speeds

Optical module selection should follow the required bandwidth, fiber infrastructure, link distance and host-port capabilities. C-LIGHT offers product options across the 100G, 200G and 400G QSFP generations.

For 400G QSFP112, the available configurations include 850nm SR4 for multimode fiber, 1310nm DR4 for 500m single-mode links, and FR4 and LR4 options for longer single-mode transmission.

When evaluating these products, customers should provide the switch or network adapter model, required transmission distance, fiber type and target optical standard. These details help establish the appropriate module configuration and host compatibility requirements.

9. Choosing the Right QSFP Generation for Network Deployment

The appropriate interface depends on the existing network architecture and the required bandwidth per port.

• Choose QSFP28 for established 100G Ethernet infrastructure, enterprise aggregation, storage networks and data center connections where 100G bandwidth is sufficient.

• Choose QSFP56 for 200G-class switching and high-performance computing networks that require higher bandwidth through four 50G-class electrical lanes.

• Choose QSFP112 for 400G platforms designed around four 100G-class electrical lanes, particularly where high port bandwidth and four-lane electrical architecture are required.

Network migration should also consider the capabilities of the switch ASIC, host SerDes, port cage, firmware and cooling system. Upgrading the transceiver alone cannot increase a port's supported bandwidth beyond the limits of the host platform.

10. Conclusion