News
Every 800G optical transceiver needs a cage to plug into, and that cage is either QSFP-DD or OSFP. The decision seems minor — both carry 8 electrical lanes of 100G PAM4, both deliver 800 Gbps. But the form factor you standardize on today determines which switches you can buy, how you handle cooling, and whether you can reuse existing infrastructure. Here is what actually separates them.
QSFP-DD (Quad Small Form-Factor Pluggable Double Density). An evolution of the QSFP family that has dominated data centers since 40G. It doubles the electrical lanes from 4 to 8 while keeping the same width as QSFP28/QSFP56. A QSFP-DD cage accepts older QSFP modules — plug a 100G QSFP28 into an 800G QSFP-DD port and it works. This backward compatibility is QSFP-DD's single biggest advantage.
OSFP (Octal Small Form-Factor Pluggable). A ground-up design for 8 lanes, slightly wider and deeper than QSFP-DD. It has an integrated heat sink that makes direct contact with the module shell — better for cooling but incompatible with QSFP cages. An OSFP port needs an adapter to accept QSFP modules, which adds cost and insertion loss.
| Parameter | QSFP-DD | OSFP |
|---|---|---|
| Dimensions (W × L × H) | 18.35 × 104 × 8.5 mm | 22.58 × 107.8 × 13 mm |
| Electrical lanes | 8 × 100G PAM4 | 8 × 100G PAM4 |
| Backward compatible | Yes — QSFP28/56 in same cage | No — adapter required |
| Thermal design | Riding heat sink (cage-level) | Integrated heat sink (module-level) |
| Max power (module) | ~20 W | ~28 W |
| Ecosystem breadth | Wider — most switch vendors | Narrower — NVIDIA/Cisco strong |
| 1.6T readiness | Yes (224G lanes in same cage) | Yes (native 224G support) |
| Best for | Mixed-generation DC fabric | Greenfield 800G, high-temp racks |
This is where most decisions are made. If you are building a greenfield 800G fabric, OSFP's thermal headroom and native 224G support make it the technically stronger choice. But if you have existing QSFP28 (100G) or QSFP56 (200G) infrastructure — and most data centers do — QSFP-DD's ability to mix generations in the same switch port is operationally transformative.
Imagine a 32-port QSFP-DD switch. You can populate 8 ports with 800G DR8 for spine uplinks, 16 ports with 400G QSFP56-DD for leaf connections, and leave 8 ports with 100G QSFP28 for management and out-of-band — all in the same switch, no adapters. OSFP cannot do this without per-port adapter modules that add $30–60 each and 0.5 dB of insertion loss.
OSFP's larger size and integrated heat sink let it handle 28 W modules — important for 1.6T coherent optics on the roadmap. QSFP-DD tops out around 20 W in most implementations. For today's 800G PAM4 modules drawing 14–18 W, either form factor works. But if you plan to run coherent ZR+ or future 1.6T optics in the same cages, OSFP's thermal margin matters.
The counterpoint: most data center hot aisles are designed around QSFP-DD thermal profiles. Switching to OSFP often means recalculating rack-level cooling — not a dealbreaker, but an engineering step many teams skip until it is too late.
Decision heuristic: If your deployment is greenfield 800G without legacy QSFP — especially for AI clusters where thermal density is the bottleneck — OSFP is the forward-looking choice. If you are upgrading an existing fabric with mixed 100G/400G/800G generations, QSFP-DD's backward compatibility saves more in operational simplicity than OSFP's thermal headroom adds in future-proofing.
APEX Group supplies 800G transceivers in both QSFP-DD and OSFP form factors — SR8, DR8, FR4, and coherent ZR+ — so network architects can match the optics to the switch platform without being locked into one form factor ecosystem.
APEX GROUP — www.apexallinone.com
