Starica logo Starica
Applications / AI Compute and GPU Clusters

AI Compute and GPU Clusters

AI Cluster Optical Interconnect: 400G and 800G Selection

Plan AI cluster optics by GPU, Leaf and Spine position, then match form factor, reach, fiber, thermal limits, FEC and breakout mode.

Selection FAQ

Scenario selection questions

Should an AI cluster use 400G or 800G?

Choose by topology, host generation and bandwidth target. Both rates may coexist across node, Leaf and Spine layers.

How should 800G reach be selected?

Evaluate VR8/SR8 for MMF, DR8 for 500m-class parallel SMF and 2xFR4 for 2km-class SMF, subject to the exact host and PMD.

What must be validated?

Validate host coding, FEC, errors, optical power, power draw, temperature, link flap, fiber mapping and long-duration stability.

Typical Scenarios

Typical deployment positions

01

GPU node uplinks

Match the NIC, switch port, rack distance and cable path.

02

Leaf-Spine fabric

Plan 400G and 800G links by reach, fiber and port density.

03

High-density breakout

Validate host breakout, lane mapping, cabling, FEC and every remote endpoint.