Wayne, PA June 03, 2025 – Cornelis Networks, an unbiased supplier of high-performance networking options, immediately launched the CN5000 household, which the corporate mentioned is the {industry}’s highest-performing end-to-end networking resolution for AI and HPC efficiency, supporting deployments with as much as 500,000 endpoints.
Engineered to deal with compute underutilization and scaling challenges, CN5000 delivers lossless knowledge switch and congestion avoidance capabilities—enhancing infrastructure efficiency and eOiciency throughout AI and HPC environments.
“Networking ought to do extra than simply transfer knowledge shortly – it ought to unlock the total potential of each compute cycle,” mentioned Lisa Spelman, CEO of Cornelis. “That’s the
efficiency we’re providing clients with the CN5000 – a brand new breed of network-led utility acceleration for AI and HPC functions the place our scale-out community
turns into a force-multiplier for efficiency at any scale.”
CN5000 permits AI and HPC functions to realize sooner and extra predictable job completion instances, and larger computational eOiciency by minimizing congestion occasions and sustaining highest throughput below load. For HPC workloads, CN5000 outperforms InfiniBand NDR with 2X greater message rates1, 35 % decrease latency1 and as much as 30 % sooner simulation times2 pushed by its superior, congestion-free structure. For AI functions, CN5000 delivers 6X sooner collective communication3 in comparison with RoCE, pushed by Omni-Path’s lossless structure.
• AI at Scale: CN5000 delivers near-linear scaling of coaching efficiency for big language fashions and unlocks superior inference with prolonged reasoning enabling higher-quality outputs and tackling beforehand intractable issues.
• Quicker, Extra Predictable HPC: CN5000 delivers as much as 30% sooner execution for compute intensive HPC workloads like computational fluid dynamics (CFD), local weather modeling, and seismic simulations with greater sustained throughput below real-world situations in comparison with InfiniBand NDR.
• Common Interoperability: CN5000 seamlessly interoperates with GPUs, CPUs, and accelerators, from AMD, Intel, NVIDIA, and extra. Totally helps OpenFabrics software program stacks for actually vendor-neutral deployments.
The CN5000 household is a complete 400G scale-out networking resolution that features:
• SuperNICs: Single and dual-port, air- and liquid-cooled
• Switches: 48-port swap and modular 576-port Director Class swap configurations, air- and liquid-cooled
• OPX Software program Suite: Open-source, industry-standard host and community administration software program
• Cabling: Passive and lively copper and lively optical choices for eOicient, high-density connectivity
This totally built-in, end-to-end resolution delivers most efficiency, scalable eOiciency, and seamless deployment throughout numerous workloads. CN5000 Product Household
Powered by the Omni-Path structure, CN5000 delivers lossless and congestion-free knowledge transmission with credit-based movement management and dynamic fine-grained adaptive
routing. Designed for max efficiency, reliability, scalability, and knowledge integrity with dynamic lane scaling and optimized hyperlink safety. The Omni-Path structure is driving superior options into the Extremely Ethernet Consortium immediately to bridge the efficiency hole between conventional networks and the calls for of scale-out computing environments.
CN5000 is step one in Cornelis’ long-term imaginative and prescient to revolutionize efficiency in scale out computing. The upcoming CN6000 (800G) will unify Omni-Path and RoCE-enabled Ethernet for double the bandwidth and seamless cross-paradigm integration. CN7000 (1.6T) is a revolutionary platform integrating Extremely Ethernet Consortium requirements with Cornelis’ confirmed Omni-Path structure, establishing a brand new efficiency frontier for probably the most demanding AI and HPC environments.
CN5000, Cornelis’ breakthrough AI and HPC scale-out networking resolution, begins delivery to clients this month and will probably be broadly accessible beginning Q3 2025 from all main OEMs.
