Comnen 1.6T OSFP-XD SR16 LPO — Linear Pluggable Optical Module

Specifications

Overview

The Comnen 1.6T OSFP-XD SR16 is a 1.6-terabit optical transceiver for short-reach AI and HPC fabric links. It carries 16 lanes of 100G PAM4 signalling over an 850nm VCSEL array to an MTP/MPO-16 APC connector, reaching 50 metres on OM4 multimode fibre.

Its distinguishing feature is the LPO architecture — Linear Pluggable Optics. A conventional 800G or 1.6T module contains a digital signal processor that retimes and reshapes the electrical signal inside the module. Linear pluggable optics remove that DSP entirely and pass the host's signal straight through to the optical engine, relying on the switch ASIC's own SerDes equalisation to close the link. The payoff is substantial: 8.5W or less for a 1.6T module, against a materially higher figure for the DSP-based equivalent, with lower latency as a second benefit.

That power saving matters at scale. In a 64-port 1.6T switch, the difference between DSP-based and linear modules is measured in hundreds of watts per chassis, which directly reduces the cooling and power distribution burden in a dense AI pod. The trade-off is that LPO requires a compatible host ASIC with sufficiently capable SerDes and a controlled link budget — which is why the module is specified with a 2 dB power budget over a 50m OM4 reach.

DDM/DOM digital diagnostics are supported for link monitoring, and the module is compliant with the OSFP-XD MSA 1.1 and CMIS 5.2 management specification. QS Compute supplies 800G and 1.6T optics, including linear pluggable variants, for AI cluster deployments.

Key Benefits

8.5W or less at 1.6T — linear pluggable optics remove the in-module DSP, cutting power and heat per port. Lower latency because there is no retiming stage in the module. MTP/MPO-16 APC with 16x100G PAM4 for straightforward 16-lane fabric mapping. Full DDM/DOM and CMIS 5.2 management for link diagnosis.

Applications

AI/ML cluster intra-rack and rack-to-rack links, HPC interconnect, 1.6T and 800G switch port breakout, data-centre spine-leaf optics, and power-constrained dense AI pods.

Related Products