Cornelis has raised $205 million in funding led by IAG Capital Partners and introduced a networking system designed to reduce the amount of time expensive AI accelerators spend waiting for data. The company says its new Active Compute Fabric allows networking hardware to process information while moving it between chips, rather than functioning only as a transport layer.
The funding will support Cornelis’ expansion into scale-up networking and development of its next generation of AI and high-performance computing products. The company spun out of Intel’s high-performance interconnect business in 2020 and has since developed networking technology based on the Omni-Path platform.
Putting More Computing Inside the Network
As AI clusters expand to thousands of GPUs and other accelerators, their performance increasingly depends on how efficiently those chips communicate. Congestion, synchronization and data transfers can leave processors idle even when substantial computing capacity remains available.
Cornelis’ Active Compute Fabric is designed to move some of that work into the network itself. The system combines lossless data transport with in-fabric acceleration and programmable computing, allowing certain communication operations to occur while information is moving through the network.
Cornelis says the approach can reduce communication overhead and keep accelerators working for a greater portion of the time. Its architecture covers both scale-out networking, which connects large numbers of servers, and scale-up systems that connect accelerators more closely within rack-scale infrastructure.
The company is also emphasizing compatibility across different hardware vendors. Rather than requiring customers to commit to a single accelerator supplier, Cornelis is building around open standards that can support chips from companies including Nvidia, AMD and Intel.
Cornelis Targets Nvidia’s Networking Position
That strategy places Cornelis against Nvidia, whose AI infrastructure business extends beyond GPUs into networking technologies including InfiniBand. Nvidia’s chips can operate with competing networking systems, but its hardware and software are closely optimized to work together, giving customers an incentive to adopt more of the company’s infrastructure stack.
Cornelis is positioning an open networking alternative for customers that want greater flexibility over their accelerator choices. In its funding and product announcement, the company said its CN5000 networking products are already shipping, while its next-generation CN6000 is being sampled by customers ahead of wider availability expected later this year.
The new capital will also support manufacturing, customer deployments and further development of Cornelis’ scale-up and scale-out networking roadmap. The company has not disclosed a new valuation alongside the funding.
Featured image credits: Magnific.com
For more stories like it, click the +Follow button at the top of this page to follow us.