{"slug": "cornelis-debuts-cn6000-to-double-networking-performance-for-ai-and-hpc", "title": "Cornelis debuts CN6000 to double networking performance for AI and HPC", "summary": "Cornelis Networks launched the CN6000 networking platform, which the company says doubles the performance of its previous generation and marks the first time Cornelis' offerings support Ethernet alongside Omni-Path. The CN6000 portfolio includes a 1U switch with 48 ports of 800 Gb/s and two superNICs: an Omni-Path version Cornelis claims supports 1.6 billion messages per second with sub-microsecond MPI latency, and an Ethernet version with RoCEv2 and Ultra Ethernet compliance. The platform is available for evaluation now, with partner systems due in the first quarter of 2027; Dell, Lenovo and Supermicro are among OEM partners planning validation, and Cornelis CEO Lisa Spelman said \"every minute an accelerator waits on the network is capacity our customers paid for and didn't use.", "body_md": "Cornelis Networks launched its next-generation CN6000 networking platform previously [teased at last year’s SC25 event](https://www.sdxcentral.com/news/cornelis-unveils-800-gbs-supernic-for-ai-and-hpc-networking/) in St. Louis, providing double the performance of the previous generation and marking the first time Cornelis' offerings support Ethernet and its traditional home of [Omni-Path](https://www.sdxcentral.com/analysis/cornelis-networks-sees-supercomputing-success-as-ethernet-vision-crystalizes/).\n\nThe CN6000 portfolio includes a network switch and \"super\" network interface cards (superNICs). The platform is currently available for evaluation, with offerings set to reach partner systems during the first quarter of 2027.\n\n“Every minute an accelerator waits on the network is capacity our customers paid for and didn’t use,” Cornelis CEO Lisa Spelman noted on the new platform. “We built CN6000 for the communication patterns that make parallel AI and HPC (high-performance computing) workloads so demanding.”\n\nTwo superNICs feature in the CN6000 line. Cornelis claims the Omni-Path iteration can support 1.6 billion messages per second and offers sub-microsecond message passing interface (MPI) latency.\n\nThe Ethernet superNIC iteration includes remote direct memory access over converged Ethernet version 2 (RoCEv2) to provide low-latency connections to Ethernet switches. It’s also Ultra Ethernet compliant (UEC) so it can be integrated into consortium-supporting Ethernet-based infrastructures.\n\nCornelis’ switch, meanwhile, is Omni-Path only and comes in a one-unit (1U) form factor. It features 48 ports of 800 Gb/s (800G) and comes packaged with adaptive routing and native multiplane support.\n\nThe underlying software ecosystem uses the same software and host driver as the previous-generation CN5000, and features support for a host of open software systems including collective communications libraries from Nvidia (NCCL) and AMD (RCCL), among others.\n\n“Omni-Path delivers the highest message rate in the industry with lossless delivery at scale, and our new Ethernet superNIC brings that same parallel computing expertise to the standards-based fabrics where most AI infrastructure runs,” Spelman added. “That combination of performance and choice is what our customers need as they scale training, inference, and scientific computing.”\n\nCornelis is looking to take its networking offerings across the industry, with the firm extending CN6000 support across accelerators from Nvidia and Qualcomm, as well as CPU platforms from AMD, Arm, and Intel. Dell, Lenovo, and Supermicro count among the OEM partners planning validation and support of CN6000.\n\nWhile acting as the custodians of Omni-Path, Spelman and company have long sought to support as many consortia as possible to open as many deployment paths for its offerings. The startup’s open vision will expand further with its forthcoming CN7000 line, roadmapped for formal unveiling in 2027, that will integrate an Omni-Path layer with UEC standards.\n\n“We're giving options and in times when standards are uncertain, and it's still being shaken out, that dual-path and multimodal capability can be extremely valuable,” Spelman [explained to *SDxCentral*](https://www.sdxcentral.com/news/cornelis-targets-inference-accelerator-utilization-with-205m-expansion-into-scale-up/) back in September.", "url": "https://wpnews.pro/news/cornelis-debuts-cn6000-to-double-networking-performance-for-ai-and-hpc", "canonical_source": "https://www.sdxcentral.com/news/cornelis-debuts-cn6000-to-double-networking-performance-for-ai-and-hpc/", "published_at": "2026-11-16 13:38:47+00:00", "updated_at": "2026-11-16 13:52:47.189622+00:00", "lang": "en", "topics": ["ai-infrastructure", "ai-chips"], "entities": ["Cornelis Networks", "CN6000", "Lisa Spelman", "Omni-Path", "Nvidia", "AMD", "Dell", "Supermicro"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/cornelis-debuts-cn6000-to-double-networking-performance-for-ai-and-hpc", "markdown": "https://wpnews.pro/news/cornelis-debuts-cn6000-to-double-networking-performance-for-ai-and-hpc.md", "text": "https://wpnews.pro/news/cornelis-debuts-cn6000-to-double-networking-performance-for-ai-and-hpc.txt", "jsonld": "https://wpnews.pro/news/cornelis-debuts-cn6000-to-double-networking-performance-for-ai-and-hpc.jsonld"}}