Networking hardware from Cornelis Networks has been selected to power the high-speed interconnect for a new computing cluster at Lawrence Livermore National Laboratory (LLNL).

Cornelis’ CN5000 will support Lynx, an upcoming supercomputing cluster that’s being used as part of the Advanced Simulation and Computing Program (ASC), which performs continual assessments and certification on the U.S. nuclear stockpile.

The ASC program was launched in the 1990s by the Department of Energy to use supercomputers to evaluate the nation's nuclear stockpile through advanced modeling and simulation, instead of live nuclear testing.

Public details on the program’s newest cluster have been limited, aside from its 952-node scale and its use of Dell Technologies hardware.

An LLNL official revealed to SDxCentral that the Lynxcluster will be powered by Dell’s C6620 and R760 servers, which feature Intel Sapphire Rapids CPUs, boasting 512 GB DDR5 memory per node. Lynx is expected to be delivered in fall 2026.

The ASC program spans LLNL, Los Alamos, and Sandia National Laboratories, with Lynx set to reside at the same facility as El Capitan, the fastest supercomputer in the world. ASCI Red, one of the first systems supporting the program, previously held that title from June 1997 to June 2000.

The CN5000 will provide scale-out networking for Lynx, with the hardware touted by Cornelis as “engineered to unlock the full performance potential of AI and HPC environments.”

Unveiled in June, Cornelis states that the CN5000 family outperforms InfiniBand NDR-based systems with two times higher message rates and up to 30% faster simulation times.

The CN5000 uses Cornelis’ Omni-Path architecture, which the firm obtained following its 2020 spin-out from Intel. Cornelis' CEO, Lisa Spelman, outlined to SDxCentral in June that the end-to-end networking platform can boost compute utilization in data center environments.

“Products that are targeted toward this space … they're getting less than 50% compute utilization, and that can depend on the cluster and all that,” Spelman said.

“But you see as low as 25%, you see lots in the 35% range, and you see best in class at a hyperscaler with a couple-hundred performance engineers getting to that 50% mark, that's a tremendous amount of compute to be left on the table.”

In addition to its compute utilization capabilities and low-latency, the CN5000 series is vendor-agnostic and supports OpenFabrics software stacks. It comes in air- and liquid-cooled SuperNICs with 48-port switches and modular 576-port Director Class switches.

Upon announcing that its networking solutions would feature in Lynx, Cornelis said it has been working with the National Nuclear Security Administration (NNSA) for the past three years as part of the Next-Generation High Performance Computing Networking project, with the pair working to develop advanced networking technologies capable of supporting future exascale supercomputers.

“With CN5000, we’re helping the DOE accelerate scientific discovery and tackle complex national security challenges,” said Rob Hays, VP of public sector at Cornelis Networks.