The Physical Layer Behind 800G Networking
The transition to 800G Ethernet represents a critical architectural leap for cloud networks and high-density AI clusters. Built on 112G PAM4 signaling per lane, 800G connectivity relies on two primary physical form factors: OSFP (Octal Small Form-Factor Pluggable) and QSFP-DD (Quad Small Form-Factor Pluggable Double Density).
Both form factors carry 8 electrical lanes at 112G PAM4 to deliver an aggregate 800G per port, but they diverge sharply in thermal design and backward compatibility.
Key Differences and Architectural Trade-Offs
Thermal Performance
OSFP features a slightly larger footprint with built-in integrated heatsink fins. This provides superior thermal dissipation (supporting up to 25W or more per module), which is essential for high-power optical transceivers and Linear Pluggable Optics (LPO). QSFP-DD maintains a standard cage size, relying on riding heatsinks to handle up to 15W to 18W per port.
Backward Compatibility
QSFP-DD is strictly backward-compatible with legacy QSFP28 (100G) and QSFP56 (400G) ports, allowing network operators to reuse existing physical infrastructure. OSFP requires mechanical adaptors to accept standard QSFP modules.
Density and Routing
Both form factors enable 32 ports per 1U switch faceplate (delivering 25.6 Tbps aggregate switch bandwidth in a 1U frame). However, OSFP-800 is increasingly favored in top-of-rack (ToR) AI switches due to its higher thermal margin during continuous high-load operations.