Cornelis Networks specialises in high-performance interconnect solutions designed to unlock the full potential of AI and High-Performance Computing (HPC) hardware. The company's core offering is an open fabric architecture that redefines the role of the network, moving it beyond passive data transport to an active compute layer. This innovative approach allows for distributed computation at every network hop, enabling functions like congestion control and collective operations to run within the network itself, rather than burdening accelerators.
The Cornelis Active Compute Fabric is engineered to deliver superior performance for demanding AI and HPC workloads. It achieves this by offering significantly higher message rates and lower latency compared to comparable 400 Gb/s solutions, such as NDR InfiniBand. This translates into better performance per networking dollar, maximising the return on investment for organisations running large-scale computational tasks. The architecture is built on decades of interconnect engineering expertise, evolving from the Omni-Path® architecture to their current CN5000 product family.
The product suite includes SuperNICs (network interface cards) with one or two 400 Gb/s ports, various switches including 48-port models and large Director Class Switches with up to 576 ports, and a range of passive copper, active copper, and active optical cables. Complementing the hardware is the OPX Software, an open-source suite covering host drivers, fabric management, and switch management, built on libfabric. This software ensures compatibility with existing MPI, OpenSHMEM, and RCCL applications without requiring code changes.
Cornelis's solutions are trusted by national laboratories, enterprise HPC and AI centres, and major OEMs and compute partners, including AMD, Dell, Intel, and Lenovo. By integrating silicon and software design, Cornelis tunes the entire fabric as a unified system, ensuring optimal performance and efficiency for the most demanding AI and HPC environments.