WAYNE, PA — Cornelis Networks has introduced the CN6000 SuperNIC, an 800-gigabit networking adapter designed to tackle the growing performance and scalability demands of artificial intelligence and high-performance computing. The company says the new adapter delivers up to 1.6 billion messages per second — the highest rate in the industry — and brings its proven Omni-Path architecture to Ethernet RoCEv2 and Ultra Ethernet environments.
As AI clusters expand toward utility-scale deployments, traditional Ethernet architectures are struggling to keep up. Cornelis CEO Lisa Spelman said the CN6000 is engineered to eliminate those constraints by offering deterministic performance, massive scalability, and broad standards support in a single platform. “One platform. Infinite scale. No compromises,” she said.
Analysts say the market is approaching a turning point. “AI clusters demand Ethernet compatibility without giving up the performance of HPC fabrics,” said Sameh Boujelbene, vice president at Dell’Oro Group. Extending Omni-Path innovations into an 800G Ethernet design, he added, addresses long-standing interoperability and scalability issues for operators building large GPU-based systems.
Engineered for Gigascale AI
The CN6000 is the first product in Cornelis’ 800G networking family and is aimed at organizations scaling large AI and simulation workloads. Key capabilities include:
- 1.6 billion messages per second and full 800 Gbps bandwidth, enabling higher GPU utilization and faster training cycles.
- Lightweight queue-pair architecture and a hardware-accelerated RoCEv2 tracking table designed to support millions of concurrent operations with predictable latency.
- Full compliance with Ultra Ethernet and RoCEv2, supported by Cornelis’ role as a major contributor to libfabric, the software interface now central to Ultra Ethernet deployments.
Industry partners view the CN6000 as an important advancement for future compute platforms. AMD said open, standards-based innovation will be critical to building next-generation infrastructure, while Intel noted that the adapter aligns with its Xeon roadmap for predictable low-latency performance at AI scale.
Lenovo, Altair, Synopsys, Atipa Technologies, Nor-Tech, Microway, PSSC Labs, and SourceCode also expressed support for the CN6000, citing needs ranging from digital-twin simulations to training clusters that demand consistent low-latency fabrics.
Cornelis will pair the SuperNIC with a new line of CN6000-series Omni-Path switches and director-class platforms, creating an end-to-end architecture for HPC and AI operators. Customer sampling is expected to begin in mid-2026, with production deployments to follow.
For the latest news on everything happening in Chester County and the surrounding area, be sure to follow MyChesCo on Google News and MSN.