aiqbench analysis
Cornelis Networks has launched the CN6000 platform, a high-performance networking solution designed to accelerate AI training and inference workloads by delivering record message rates at 800G speeds. This hardware addresses critical infrastructure bottlenecks for large-scale models by offering low-latency RoCEv2 on standard Ethernet switches and supporting major accelerators from NVIDIA, AMD, and Intel. The release represents a significant step in optimizing data center fabrics to handle the growing demands of disaggregated inference and KV-cache transfers.
Cornelis CN6000 Networking Platform Delivers Industry-leading 1.6 Billion Messages per Second at 800G for AI and HPC - The Future of High Performance Fabrics

From Cornelis Networks
ExcerptBrings more than a decade of supercomputer-scale networking and consecutive first-silicon successes to mainstream AI; new Ethernet SuperNIC adds low-latency RoCEv2 to standard 800G switches, with Ultra Ethernet to follow.
Omni-Path SuperNIC: Industry’s highest message rate at 800G. The CN6000 Omni-Path SuperNIC doubles bandwidth to 800G and delivers 1.6 billion messages per second, the industry’s highest, with sub-microsecond MPI latency 21% lower than NVIDIA InfiniBand XDR at 800G.¹
New Ethernet SuperNIC: Low-latency RoCEv2 on standard Ethernet. The CN6000 Ethernet SuperNIC brings hardware-accelerated RoCEv2 to industry-standard 800G Ethernet switches, for latency-sensitive AI traffic such as KV-cache transfers in disaggregated inference. Ultra Ethernet is supported through software updates.
Broad silicon and system ecosystem: CN6000 supports AMD, Arm-based and Intel CPUs as well as AMD, NVIDIA and Qualcomm accelerators. OEM and system partners planning availability include Dell Technologies, Hyve Solutions, Lenovo, NEC, Penguin Solutions, Supermicro and more.
Excerpt shown under the publisher's RSS terms · 144 words · full text and figures are on the publisher's site
Other coverage of this story
- Cornelis Networks · this article
- HPCwire · Cornelis CN6000 Networking Platform Delivers 1.6B Messages per Second at 800G for AI and HPC