Cornelis Networks has recently secured $205 million to advance its innovative approach to networking in the context of artificial intelligence (AI) and high-performance computing (HPC). The company aims to alleviate data processing bottlenecks by integrating programmable compute capabilities within the network.
Developing from its roots at Intel in 2020, when the team behind the Omni-Path architecture became an independent entity, Cornelis has focused on creating a new architecture called Active Compute Fabric. This enables networking hardware to do more than just transfer data; it can now engage in computation, addressing the inefficiencies associated with traditional networking solutions which often leave computing resources underutilized.
Cornelis initially built its first products around the proprietary Omni-Path architecture, designed exclusively to connect specific networking hardware. However, with the release of the new CN6000 SuperNIC, the company has added support for Ethernet, allowing for compatibility with other vendors while maintaining its distinct architecture.
The CN6000 is notable for its multimode silicon, which can carry out both Omni-Path and Ethernet protocols. This flexible approach retains key advantages even when operating on Ethernet standards, such as efficient traffic management and reduced latency.
One of the standout features of Cornelis Networks is the Active Compute Fabric, which introduces compute processes directly into networking hardware. This includes incorporating RISC-V cores within both network interface cards (NICs) and switches, enabling them to perform AI tasks without relying heavily on general-purpose GPU cycles. This innovation aims to increase GPU utilization significantly, addressing the challenge that many organizations face with underused computing resources.
The company also acknowledges the broader market, extending its offerings beyond large-scale data centers to enterprises, governments, and academic institutions. By doing so, Cornelis seeks to optimize AI processing capabilities across various sectors that are looking to build their own AI infrastructures without being heavily dependent on the Nvidia ecosystem.
In summary, Cornelis Networks is leading the charge in enhancing AI efficiency and overall network performance through its groundbreaking approach of embedding compute power directly into networking systems, thus transforming the landscape of data-intensive applications.
