Cornelis Networks has introduced the CN5000 product family, a 400G end-to-end networking platform designed to address the performance bottlenecks in AI and high-performance computing (HPC) workloads. The launch marks a significant milestone for the company, which spun out of Intel’s Omni-Path division, and is now positioning itself as a key provider of intelligent fabric solutions for massively scaled GPU clusters.
At the heart of the CN5000 platform is the next-generation Omni-Path architecture, which delivers lossless data transmission and congestion avoidance using credit-based flow control and adaptive routing. According to Cornelis, the CN5000 outperforms comparable networks by delivering 2X greater message rates, 35% lower latency, and up to 30% higher HPC application performance compared to InfiniBand NDR. For AI environments, the CN5000 is optimized for collective operations, delivering 6X faster performance than RoCE-based systems.
“Networking should do more than just move data quickly — it should unlock the full potential of every compute cycle,” said Lisa Spelman, CEO of Cornelis Networks. “That’s the performance we are offering customers with the CN5000 — a new breed of network-led application acceleration for AI and HPC applications where our scale-out network becomes a force-multiplier for performance at any scale.”
Key Technical Features of CN5000:
- 400G Scale-Out Fabric supporting deployments of up to 500,000 endpoints
- SuperNICs: Single and dual-port options with air and liquid cooling
- Switches: 48-port units and modular Director-class configurations with up to 576 ports
- Congestion Management: Lossless transport with credit-based flow control and adaptive routing
- OPX Software Suite: Built on open-source frameworks for vendor-neutral deployment
- Universal Interoperability: Supports AMD, Intel, NVIDIA GPUs and CPUs
Looking ahead, Cornelis is already preparing the CN6000 series, an 800G offering that will integrate Omni-Path capabilities with RoCE-enabled Ethernet to target broader cloud and enterprise markets. The company also plans to launch the CN7000 (1.6T) platform aligned with Ultra Ethernet Consortium standards to meet the escalating demands of AI and exascale systems.
