Ultra-high bandwidth & dedicated chip: Featuring Broadcom Tomahawk6 chip and 204.8T ultra-large bandwidth, the switch delivers full wire-speed on all 128×800G OSFP112 ports. Only two-layer networking enables a 60,000-GPU cluster, cutting switch quantity by 40% and drastically lowering cluster deployment complexity and Total Cost of Ownership (TCO).

Jitter-free intelligent computing network: It supports RoCEv2 lossless communication with dynamic load balancing and zero-packet-loss architecture to guarantee stable operation of AI training workloads.

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

Multi-rate backward compatibility: OSFP112 ports support 400G/200G via beam splitting technology to match networking requirements of diversified computing nodes.

Full-link redundancy: Adopts 2+2 redundant 2700W CRPS power supplies and 4+1 redundant fans with military-grade reliability design to sustain 7×24-hour heavy-load continuous operation.

Optimized energy efficiency: Low power consumption realized through intelligent temperature control and adaptive energy-saving algorithms; robust thermal solution consists of customized VC vapor chamber coolers and superconductive cooling fins.