Ultra-high bandwidth with dedicated chip: Equipped with Broadcom Tomahawk5 chip and delivering 102.4T ultra-high bandwidth, the switch features 128×400G QSFP112 full wire-speed ports to fully satisfy high-throughput requirements of ten-thousand-card AI clusters.

Jitter-free intelligent computing network: Supports lossless RoCEv2 communication alongside dynamic load balancing and zero-packet-loss design to ensure stable execution of AI training jobs.

Open ecosystem compatibility: Compatible with Open Networking Interface (ONI) and Sonics/NOS software to balance flexible deployment and ecosystem adaptability.

Multi-rate auto-adaptation: QSFP112 ports are backward compatible with 200G/100G/25G via beam splitting technology, matching network specifications of various computing nodes.

Full-link redundancy: Adopts 2+2 redundant 1600W power supplies and N+1 redundant fans with military-grade reliable construction for 7×24-hour heavy-duty continuous operation.

High-efficiency power design: Low power consumption enabled by intelligent temperature control and adaptive energy-saving algorithms; robust thermal solution with customized VC vapor chamber coolers and superconductive cooling fins.