\nFeature<\/b><\/strong><\/td>\nQSFP28<\/b><\/strong><\/td>\nQSFP56<\/b><\/strong><\/td>\n<\/tr>\n\nData rate<\/strong><\/td>\n| 100 Gbps<\/td>\n | 200 Gbps<\/td>\n<\/tr>\n | \nElectrical lanes<\/strong><\/td>\n| 4 x 25 Gbps<\/td>\n | 4 x ~50 Gbps<\/td>\n<\/tr>\n | \nModulation<\/strong><\/td>\n| NRZ (1 bit\/symbol)<\/td>\n | PAM4 (2 bits\/symbol)<\/td>\n<\/tr>\n | \nElectrical interface<\/strong><\/td>\n| 100GAUI-4<\/td>\n | 200GAUI-4<\/td>\n<\/tr>\n | \nForm factor<\/strong><\/td>\n| QSFP<\/td>\n | QSFP (same size)<\/td>\n<\/tr>\n | \nTypical power<\/strong><\/td>\n| 3.5 W to 5.5 W<\/td>\n | 4.5 W to 7.5 W<\/td>\n<\/tr>\n | \nCommon reach<\/strong><\/td>\n| SR4 100 m, LR4 10 km<\/td>\n | SR4 100 m, FR4 2 km, LR4 10 km<\/td>\n<\/tr>\n | \nBackward compatibility<\/strong><\/td>\n| Works in QSFP56-capable ports with host support<\/td>\n | Does not work in QSFP28-only ports<\/td>\n<\/tr>\n | \nPrimary standards<\/strong><\/td>\n| IEEE 802.3bm, QSFP28 MSA<\/td>\n | IEEE 802.3bs, IEEE 802.3cd, QSFP56 MSA<\/td>\n<\/tr>\n | \nTypical use case<\/strong><\/td>\n| 100G leaf-spine, mature fabrics<\/td>\n | 200G spine, AI\/ML clusters, 400G migration<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n <\/p>\n Data Rate and Lane Architecture<\/strong><\/h3>\nQSFP28 delivers 100 Gbps across four 25 Gbps NRZ lanes. QSFP56 doubles that to 200 Gbps by running four ~50 Gbps PAM4 lanes in the same four-lane structure. The physical connector and cage are unchanged, but the electrical signaling is not interchangeable.<\/p>\n <\/p>\n Modulation: NRZ vs PAM4<\/strong><\/h3>\nNRZ uses two voltage levels to represent 0 and 1. PAM4 uses four voltage levels to represent 00, 01, 10, and 11, effectively transmitting two bits per symbol. This doubling comes at the cost of reduced signal-to-noise margin and stricter forward error correction (FEC) requirements. We explore this in more detail in the next section.<\/p>\n <\/p>\n Power Consumption and Thermal Design<\/strong><\/h3>\nQSFP56 modules consume more power than QSFP28 modules because of the higher-speed DSP and FEC logic required for PAM4. A typical QSFP28 module draws 3.5 W to 5.5 W, while a QSFP56 module draws 4.5 W to 7.5 W. In a fully populated 32-port switch, that difference can add 32 W to 64 W per switch. For hyperscale deployments with thousands of ports, thermal and power budgets matter.<\/p>\n <\/p>\n Backward and Forward Compatibility<\/strong><\/h3>\nThe same QSFP cage accepts both modules, but electrical compatibility is directional:<\/p>\n \n- \u2022<\/span><\/strong>A QSFP28 module can often operate in a QSFP56-capable port if the host supports 25G NRZ fallback.<\/li>\n
- \u2022<\/span><\/strong>A QSFP56 module cannot operate in a QSFP28-only port because the host lacks 50G PAM4 SerDes.<\/li>\n
- \u2022<\/span><\/strong>QSFP56 modules can plug into QSFP-DD ports that support 200G electrical modes, providing a forward path.<\/li>\n<\/ul>\n
Always verify the switch datasheet, ASIC generation, firmware version, and port configuration before assuming interoperability.<\/p>\n <\/p>\n Cost and Cost-per-Gbps<\/strong><\/h3>\nThe sticker price of a QSFP56 module is higher than a comparable QSFP28 module, but the cost-per-Gbps is often lower. For example, replacing a $250 100GBASE-SR4 QSFP28 module with a $300 200GBASE-SR4 QSFP56 module nearly doubles bandwidth while increasing module cost by only about 20%. That translates to roughly 40% lower cost-per-Gbps, assuming the switch platform can use the extra capacity.<\/p>\n However, the total cost of ownership also includes switch ASIC upgrades, higher power draw, cooling, and potentially new cables. The economics favor QSFP56 when you need the bandwidth and can absorb the incremental power.<\/p>\n <\/p>\n <\/p>\n NRZ vs PAM4: Why Modulation Defines the Difference<\/strong><\/h2>\nThe most important technical distinction between QSFP28 and QSFP56 is not the form factor. It is the modulation.<\/p>\n <\/p>\n NRZ Encoding in QSFP28<\/strong><\/h3>\nNRZ is a binary modulation scheme. Each symbol carries one bit. The signal transitions between two voltage levels, representing 0 and 1. NRZ is simple, has good noise immunity, and has been the workhorse of 10G, 25G, and 100G Ethernet.<\/p>\n <\/p>\n PAM4 Encoding in QSFP56<\/strong><\/h3>\nPAM4 uses four voltage levels to encode two bits per symbol. At the same baud rate, PAM4 carries twice as much data as NRZ. This is how QSFP56 achieves 50 Gbps per lane without doubling the clock rate. The trade-off is a smaller voltage eye between levels, which makes the signal more susceptible to noise and crosstalk.<\/p>\n <\/p>\n FEC, DSP, and Signal Integrity<\/strong><\/h3>\nBecause PAM4 has a lower noise margin, reliable operation requires stronger FEC and digital signal processing (DSP) in both the switch ASIC and the optical module. IEEE 802.3bs specifies Reed-Solomon FEC (RS-FEC) for 200GBASE optical links. This adds a small amount of latency but enables error-free transmission over standard fiber.<\/p>\n Cable quality, connector cleanliness, and PCB trace design also become more critical with PAM4. A marginal link that would work fine with NRZ may fail with PAM4 unless all components in the channel meet specifications.<\/p>\n <\/p>\n Practical Impact on Network Design<\/strong><\/h3>\nWhen you move from QSFP28 to QSFP56, you are not just buying faster optics. You are committing to:<\/p>\n \n- \u2022<\/span><\/strong>PAM4-capable switch ASICs<\/li>\n
- \u2022<\/span><\/strong>Stronger FEC configuration<\/li>\n
- \u2022<\/span><\/strong>Tighter signal-integrity requirements<\/li>\n
- \u2022<\/span><\/strong>Higher power and thermal budgets<\/li>\n<\/ul>\n
This is why many engineers treat QSFP56 as a platform decision, not just a module upgrade.<\/p>\n <\/p>\n ![\"NRZ](\"https:\/\/ascentoptics.com\/blog\/wp-content\/uploads\/2026\/06\/NRZ-vs-PAM4-Modulation-Diagram.png\") <\/p>\n
<\/p>\n <\/p>\n Compatibility and Interoperability<\/strong><\/h2>\nCompatibility is the area where most mistakes happen. The modules share the same mechanical envelope, but the electrical interface is what determines whether a link comes up.<\/p>\n <\/p>\n Can QSFP28 Work in a QSFP56 Port?<\/strong><\/h3>\nOften, yes. Many QSFP56-capable switch ports support speed fallback to 25G NRZ per lane, allowing a QSFP28 module to run at 100 Gbps. This depends on the switch ASIC, firmware, and vendor configuration. For example, some Arista 7060X4 and NVIDIA Spectrum-2\/3 platforms support this mode. Check the vendor compatibility matrix before relying on it.<\/p>\n <\/p>\n Can QSFP56 Work in a QSFP28 Port?<\/strong><\/h3>\nNo. A QSFP28-only port uses 25G NRZ SerDes and cannot interpret the 50G PAM4 signaling from a QSFP56 module. The module may physically fit, but the link will not initialize.<\/p>\n <\/p>\n Major Switch Platform Support<\/strong><\/h3>\nThe following platforms are examples of hardware that support QSFP56; always confirm the specific SKU and software release:<\/p>\n \n- \u2022<\/span>Cisco<\/strong>: selected Nexus 9000 series switches with QSFP56-capable line cards<\/li>\n
- \u2022<\/span>Arista<\/strong>: 7060X4, 7060X6, and 7170 series<\/li>\n
- \u2022<\/span>Juniper<\/strong>: QFX5200, QFX5220, and PTX platforms<\/li>\n
- \u2022<\/span>NVIDIA\/Mellanox<\/strong>: Spectrum-2 (SN3700) and Spectrum-3 (SN4600) switches<\/li>\n
- \u2022<\/span>Huawei<\/strong>: CloudEngine 9800, 8800, and 6800 series with 200GE interfaces<\/li>\n<\/ul>\n
<\/p>\n DAC and AOC Cable Compatibility<\/strong><\/h3>\nQSFP28 DACs and AOCs are generally not suitable for 200G QSFP56 links because the cable wiring and signal conditioning are designed for 25G NRZ per lane, not 50G PAM4. Some switches can down-rate a QSFP56 port to 100G and use a QSFP28 cable, but this is a host-dependent feature.<\/p>\n For native 200G links, use QSFP56-rated DACs or AOCs. For longer reaches, use the appropriate optical transceiver module matched to your fiber type and distance.<\/p>\n <\/p>\n ![\"QSFP28](\"https:\/\/ascentoptics.com\/blog\/wp-content\/uploads\/2026\/06\/Compatibility-Matrix.png\") <\/p>\n
<\/p>\n <\/p>\n When to Choose QSFP28 vs QSFP56<\/strong><\/h2>\nThe right transceiver depends on bandwidth needs, existing hardware, budget, and migration plans.<\/p>\n <\/p>\n Choose QSFP28 If<\/strong><\/h3>\n\n- \u2022<\/span><\/strong>Your network requires 100 Gbps per port today<\/li>\n
- \u2022<\/span><\/strong>You are extending an existing 100G fabric with mature, low-cost optics<\/li>\n
- \u2022<\/span><\/strong>Power and cooling budgets are constrained<\/li>\n
- \u2022<\/span><\/strong>Your switches are QSFP28-only and cannot be upgraded<\/li>\n
- \u2022<\/span><\/strong>You need maximum compatibility across vendor platforms<\/li>\n<\/ul>\n
<\/p>\n Choose QSFP56 If<\/strong><\/h3>\n\n- \u2022<\/span><\/strong>Your spine layer needs 200 Gbps per port<\/li>\n
- \u2022<\/span><\/strong>You are building AI\/ML or high-performance computing fabrics<\/li>\n
- \u2022<\/span><\/strong>Your switches support QSFP56 with 50G PAM4 SerDes<\/li>\n
- \u2022<\/span><\/strong>You want a lower cost-per-Gbps for high-bandwidth links<\/li>\n
- \u2022<\/span><\/strong>You plan to migrate to 400G QSFP-DD in the same chassis later<\/li>\n<\/ul>\n
<\/p>\n Skip Both and Go Directly to QSFP-DD or OSFP?<\/strong><\/h3>\nSometimes yes. If you are deploying a new fabric and expect to need 400G within 12 to 24 months, choosing a QSFP-DD or OSFP switch platform from the start may be more cost-effective than an intermediate 200G step. These newer form factors support 400G and 800G using eight lanes, and many QSFP-DD ports can accept QSFP56 modules for 200G operation.<\/p>\n <\/p>\n <\/p>\n Migration and Future-Proofing<\/strong><\/h2>\nMoving from 100G to 200G is rarely a forklift upgrade. Most organizations phase it in based on traffic growth and hardware refresh cycles.<\/p>\n <\/p>\n QSFP28 to QSFP56 Upgrade Path<\/strong><\/h3>\nA common approach is to deploy QSFP56-capable switches only in the spine layer while keeping 100G leaf switches. This doubles spine capacity without touching the leaf layer. As leaf switches are refreshed, they can also move to 200G or remain at 100G with breakout cables.<\/p>\n <\/p>\n QSFP56 as a Bridge to 400G<\/strong><\/h3>\nQSFP56 is often described as a bridge technology. It delivers 200G today and can be reused in QSFP-DD ports that support 200G electrical modes. When those ports are later upgraded to 400G, the QSFP56 modules can be redeployed elsewhere or phased out.<\/p>\n <\/p>\n Rack-Level Density and Power Planning<\/strong><\/h3>\nA 32-port QSFP56 switch can deliver 6.4 Tbps of switching capacity in 1 RU. That density is attractive, but it also concentrates heat. Plan for:<\/p>\n \n- \u2022<\/span><\/strong>Power consumption per module (5 W to 8 W typical)<\/li>\n
- \u2022<\/span><\/strong>Total switch power and thermal design power (TDP)<\/li>\n
- \u2022<\/span><\/strong>Front-to-back or back-to-front airflow alignment<\/li>\n
| | | | | | | | | | | | | |
![\"QSFP28](\"https:\/\/ascentoptics.com\/blog\/wp-content\/uploads\/2026\/06\/QSFP-Lane-Architecture-Comparison.png\")
<\/p>\n