Cornelis Unveils AI Network Design Built for AMD Platforms

Cornelis Networks

WAYNE, PA — Cornelis introduced a reference architecture combining its next-generation CN6000 SuperNIC with upcoming AMD processors and AI accelerators, expanding a five-year partnership aimed at high-performance computing and large-scale artificial intelligence deployments.

The design pairs the CN6000 networking platform with AMD’s 6th Gen EPYC processors and Instinct MI400 series GPUs, providing infrastructure operators with a blueprint for building AI training, inference and high-performance computing (HPC) systems. The company expects the CN6000 family to become generally available during the fourth quarter of calendar year 2026.

The announcement reflects growing demand for networking hardware capable of matching the performance of increasingly powerful AI processors, as large language models and HPC workloads require faster communication between thousands of computing nodes.

READ:  McCormick Presses AI Buildout With Worker, Data Center Rules

Cornelis designed the CN6000 to support multiple networking protocols—including Omni-Path, RoCEv2 and Ultra Ethernet—through a single network adapter. The SuperNIC is engineered to deliver 800 gigabits per second of bandwidth through a PCIe 6.0 x16 interface and is targeting more than 1.6 billion bidirectional messages per second.

The company reported that pre-production modeling found the architecture completed AllReduce communication operations approximately 24% faster than conventional Ethernet configurations. Simulations also indicated training time for a 250-billion-parameter AI model on a 10,000-GPU cluster could be reduced by about 13%.

READ:  Interior Directs $110 Million to Federal AI Science Push

Cornelis stated the same networking advantages are intended to improve AI inference workloads, where rapid communication between processing stages is essential to maintaining throughput as request volumes increase.

The CN6000 platform combines the SuperNIC with a 48-port Omni-Path switch and a libfabric-based software stack designed to work with AMD’s ROCm software platform.

The reference architecture builds on a collaboration between Cornelis and AMD that spans more than five years and includes over 25 validated platform configurations, nine original equipment manufacturing partners and a joint HPC Center of Excellence in Munich.

READ:  Transportation Department Launches AI Initiative to Modernize Infrastructure

Lisa Spelman, chief executive officer of Cornelis, indicated that the collaboration is intended to provide infrastructure operators with an alternative approach for building AI computing environments.

“As AMD gets ready to share more of its AI vision and roadmap, Cornelis is proud to be one of the partners building alongside AMD for our joint customers,” Spelman said.

Support the local news that supports Chester County. MyChesCo delivers reliable, fact-based reporting and essential community resources—free for everyone. If you value that, click here to become a patron today.