Cornelis Networks CN5000 Omni-Path — 400G Lossless AI & HPC Scale-Out Fabric

Specifications

Overview

The Cornelis Networks CN5000 Omni-Path is a 400 Gb/s end-to-end fabric engineered specifically for AI and HPC clusters at scale. Rather than assembling a network from independently designed parts, CN5000 ships as a complete fabric — silicon, adapters, switches, software and cabling engineered as a single architecture and optimized to work together.

For GPU-dense AI clusters the limiting factor is rarely raw bandwidth but sustained, predictable delivery under load. CN5000 combines credit-based flow control, enhanced congestion management and fine-grained adaptive routing to keep tail latency controlled as traffic scales, so accelerators stay fed instead of waiting on the interconnect.

The family spans low-profile PCIe Gen 5 x16 SuperNICs, a 48-port 400 Gb/s leaf switch, and a 230.4 Tb/s Director Class Switch that scales to 576 ports — with air and liquid cooling options so the same fabric fits both traditional and high-density AI data center designs.

Key Benefits

Lossless at scale: congestion control and adaptive routing hold performance steady as endpoints grow toward 500,000. Full-stack design: NICs, switches, cables and software developed as one system, reaching ≈96% of line rate on 400 Gb/s. Deployment flexibility: air- or liquid-cooled switches, 400G or dual-200G port modes, and an open, standards-based software stack.

Applications

Large-scale AI model training and inference clusters · HPC simulation and modeling · GPU and accelerator scale-out fabrics · AI factory and research HPC data center builds

Related Products