Skip to content

Cornelis Networks Wins $18M NNSA Award to Advance Omni-Path

Intel spinoff Cornelis Networks, custodian and developer of the Omni-Path networking portfolio, is now closer to reaching its next-gen networking roadmap targets thanks to an R&D contract with the Department of Energy’s National Nuclear Security Administration (NNSA). The contract is valued at $18 million.

The Next-Generation High Performance Computing Network (NG-HPCN) project brings together NNSA labs and Cornelis to design and productize next-generation interconnect technologies in support of NNSA’s scientific and engineering workloads and the high-performance computing space more broadly. The project is overseen by Lawrence Livermore National Laboratory (LLNL) for the NNSA Tri-Labs: Lawrence Livermore, Los Alamos and Sandia National Laboratories.

“The HPC network is a very critical part of our supercomputers,” Matt Leininger, senior principal HPC Strategist at LLNL, told HPCwire. “It’s one of the most critical elements to distinguish our platforms from other types of computing platforms. We need the networking space to not only meet our goals, but to be highly competitive across the US marketplace and allow us to develop best-of-breed solutions. In general, a very competitive marketplace for high performance network technology is critical to not only where we are today, but where we need to be moving into the post-exascale era as well.”

Cornelis said the funding – along with other investments – will cover software and hardware R&D and will put the company on a path to launching its 400Gbps product (OPA-400) next year. When Cornelis Networks took over the Omni-Path product line from Intel in 2020, it put out a new roadmap that skipped Omni-Path 200, but introduced Omni-Path Express (OPX), which Cornelis told us is slated for general availability in a few weeks. As we reported last year, OPX is powered by an optimized host software that supports OpenFabrics Interfaces (OFI), developed under the OpenFabrics Alliance.

Cornelis emphasized three target areas for the NNSA funding: enabling open-source host software based on OpenFabrics Interfaces, accelerating network simulation capabilities for the forthcoming Omni-Path 400 Gbps product line, and development of the company’s next-generation switching infrastructure. “We’re working closely with Livermore to research what the best way is to manage traffic, using techniques like dynamic adaptive routing and congestion control architecture,” said Phil Murphy, Cornelis co-founder and chief executive officer.

LLNL’s Ruby supercomputer supports nuclear security and Covid-19 research

Cornelis and Livermore already have an established partnership. Across the Tri-Lab complex, Leininger estimates there are some 15,000 to 20,000 nodes that are using Intel/Cornelis Omni-Path networking technology, comprising a number of clusters, among them “Ruby” and “Magma.” And last fall, Cornelis Networks was selected as the primary networking technology partner for the NNSA’s Tri-Lab CTS-2 contract. Livermore will start taking delivery of CTS-2 systems with Cornelis OPA-100 networking later this year and will have the option to procure OPA-400 products as they become available under CTS-2 as well. The lab is also testing Cornelis’ OPX software ahead of its general availability.

The R&D contract announced today is being funded by the NNSA’s Advanced Simulation Computing (ASC) program as part of a post-Exascale Computing Initiative (ECI) investment portfolio intended to continue the momentum of previous ECI programs, such as the DOE’s PathForward program. That program provided funding to six technology vendors (among them Intel, the creator of Omni-Path) in support of meeting US exascale computing milestones.

As with previous exascale-focused funding projects, the goal is for the co-design collaboration with Cornelis to have an impact beyond the DOE and NNSA missions.

“We certainly expect that any of the benefits that come out of this will not only benefit the programs at the ASC program at our three laboratories, but [also] anyone doing large HPC simulations, as well as data analytics and various AI and machine learning spaces as well. We want the impact to be as broad as possible,” said Leininger.

Leave a Reply

Your email address will not be published.