SFP28 vs SFP112: 25G NRZ and 100G PAM4 Optical Interface Selection Guide

SFP28 and SFP112 share the SFP family form factor lineage, but they target fundamentally different speed tiers: SFP28 carries 25Gbps using NRZ, while SFP112 delivers 100Gbps on a single lane using PAM4. The jump from 25G to 100G changes modulation format, power profile, host requirements, and compatibility rules. Selection should be driven by port capability, service rate, thermal budget, and whether backward compatibility with existing SFP28 infrastructure is required.

1. What Is SFP28?

SFP28 is a single-lane 25Gbps optical module form factor built on the SFP+ physical footprint. It uses NRZ modulation and typically operates over one electrical lane at 28Gbps line rate. SFP28 became the mainstream 25G interface for data center server access, Top-of-Rack switching, and 5G fronthaul, where its balance of speed, power, and cost made it a practical replacement for 10G SFP+ in high-density environments.

SFP28 modules generally consume between roughly 1W and 2.5W depending on reach and optics type. Short-reach multimode and AOC variants sit at the low end, while longer-reach single-mode and DWDM variants approach the upper end. The form factor is backward compatible with SFP+ cages, so an SFP28 port can often accept an SFP+ module at 10G, but an SFP+ host generally cannot run an SFP28 module at 25G.[reference:0]

2. What Is SFP112?

SFP112 is the 100Gbps member of the SFP form factor family. It uses a single electrical lane and PAM4 modulation at approximately 53Gbaud to achieve 100G per module. The “112” in the name refers to the 112Gbps lane rate in the SFF naming convention, where PAM4 is assumed at 112G and above. SFP112 is positioned as a compact, single-lane 100G alternative to four-lane QSFP28, enabling 100G on a port density and physical footprint closer to SFP28.[reference:1]

SFP112 modules typically consume more power than SFP28, often in the range of roughly 2W to 4W depending on reach and implementation. Short-reach optics can sit near 2W, while longer-reach ER variants approach 3.5W or slightly above. The form factor is backward compatible at the cage and connector level with SFP+/SFP28, meaning an SFP112 host port can generally accept SFP28 or SFP+ modules and operate them at their native lower speeds.[reference:2][reference:3]

3. SFP28 vs SFP112: Key Specifications

4. Compatibility and the Direction of Backward Support

SFP112 also introduces a new electrical specification for the high-speed lane, because 100G PAM4 places different demands on the connector and signal path than 25G NRZ. The management interface typically uses CMIS, and the host must support the appropriate FEC and link training for 100G operation. Deploying SFP112 on a port that was designed only for SFP28 will generally not yield 100G and may not link at all.[reference:6]

5. Power, Thermal, and Port Density

6. Application Fit

6.1 SFP28: Mature 25G Access and Fronthaul

SFP28 remains the workhorse 25G interface for data center server access, Top-of-Rack switching, enterprise aggregation, and 5G fronthaul. Its NRZ signaling, low power, and broad optics portfolio make it a stable choice for large-scale 25G fabrics and radio access networks where 25G per port matches the traffic profile.[reference:7]

6.2 SFP112: Single-Lane 100G for AI and High-Density Interconnect

SFP112 is aimed at 100G links that benefit from a compact, single-lane form factor. Typical deployments include 100G server-to-switch interconnects, AI/GPU cluster networking, 100G breakout from higher-speed switch ports, and next-generation aggregation where front-panel density and cabling simplicity matter. It allows 100G to be delivered on a port that is physically close to SFP28 in size, bridging the gap between 25G access and four-lane 100G aggregation.[reference:8][reference:9]

6.3 Mixed 25G and 100G Design

A practical network can use SFP28 for 25G access and server ports while deploying SFP112 for 100G uplinks or AI cluster connections. In this design, SFP112 ports can host SFP28 modules at 25G, preserving investment in existing 25G optics, while new 100G links use SFP112 modules. The host port must be provisioned and validated for both modes.

7. Selection Guidance

Start with the required port speed and the host’s signaling capabilities. If the requirement is 25G and the host supports NRZ 25G, SFP28 offers a mature, power-efficient solution with a wide optics selection. If the requirement is 100G and the host supports PAM4 100G on a single lane, SFP112 provides a compact alternative to QSFP28 with cage-level backward compatibility to SFP28.

Verify FEC support, management interface (CMIS), thermal budget, and the vendor’s compatibility list. Do not assume that an SFP112 module will deliver 100G in an SFP28-only host, and do not assume that a 100G-capable SFP112 port will run 100G without the correct host configuration and optics pairing.

8. Frequently Asked Questions

9. Conclusion

SFP28 and SFP112 occupy different points on the SFP speed curve. SFP28 is the mature 25G NRZ interface for access, fronthaul, and cost-sensitive high-density 25G ports. SFP112 brings 100G PAM4 into the SFP family on a single lane, enabling compact 100G interconnects and AI cluster networking while retaining cage-level backward compatibility with SFP28. The right choice depends on the host’s signaling capability, the required per-port speed, thermal headroom, and the optics portfolio available for the target reach.