Intel’s Omni-Path Architecture high-speed interconnect technology is getting a second life under a stand-alone company that has spun out from the chipmaker and raised US$20 million in venture capital, giving partners once again an alternative to Mellanox’s InfiniBand that is now owned by Nvidia.
The new company, Cornelis Networks, came out of stealth mode and announced its Series A funding round Wednesday, saying that the financing was led by Downing Ventures, a division of London-based investment firm Downing LLP that focuses on early stage startups. The round’s other investors were Chestnut Street Ventures and Intel Capital, the chipmaker’s venture arm.
Cornelis was founded by former Intel employees Philip A. Murphy Jr., Vladimir K. Tamarkin and Gunnar K. Gunnarson, who had all previously worked on Omni-Path under Intel, and the company has 50 employees, 40 of which were hired from Intel. The company is based in Wayne, Pa., an area where some of Intel’s Omni-Path team previously worked.
The spin-out of Omni-Path—which involved a transfer of intellectual property, product inventory, support agreements and distribution agreements in exchange for Intel Capital receiving a stake in the company, according to Murphy—marks a reversal of sorts for the technology and product line.
CRN reported more than a year ago that Intel had halted development for its second-generation, 200-Gbps version of Omni-Path Architecture while pledging that it would continue to sell and support the first-generation, 100-Gbps products, giving partners little incentive to continue investing in the technology. The move was a boon for Mellanox Technologies, which was already shipping a 200-Gbps version of its InfiniBand interconnect and was in the process of getting acquired by Nvidia.
In an interview with CRN, Murphy, one of Omni-Path’s architects and CEO of Cornelis, said he and his colleagues began discussing with Intel the possibility of spinning out Omni-Path as a stand-alone company about 18 months ago when it was becoming clear that Intel was “de-emphasizing” the technology ahead of any official plans to discontinue future development.
Murphy had started working on Omni-Path after he came to Intel through its 2012 acquisition of the InfiniBand technology from QLogic, which had previously acquired a company he founded in 2000 that was working on an InfiniBand-based interconnect for high-performance computing. And because of all that experience, he was intimately familiar with Omni-Path and knew it could still make an impact.
“It started becoming apparent to me that Intel was going to de-emphasize this product line. And I spent the bulk of my career working on it, and I could see that it still had very big potential,” he said. “And so I approached them quite a while ago and said, ‘You know, I don’t think this fits in Intel anymore. I still think it’s fantastic technology, and I think that if we set it free that it can really make a big impact. And fortunately, Intel agreed.”
With the funding round, Murphy said Cornelis has three main focuses: supporting the more than 500 end-user sites that currently use first-generation Omni-Path products; continuing to sell the first-generation products through channel partners and distributors; and developing next-generation products, which could take on a new brand name.
Among Omni-Path’s current customer base are several high-performance computing sites, including the U.S. Department of Energy’s Lawrence Livermore National Laboratory.
In a statement, Intel said Cornelis will be “an important ecosystem partner for Intel” that will “expand the choice of scale-out interconnects for our customers.”
“Intel believes Cornelis Networks will fill the current gap and expand the ecosystem of high-performance fabric solutions, offering an option to customers building clusters for HPC and AI based on Intel Xeon processors,” Trish Damkroger, vice president and general manager of Intel’s high-performance computing group, said in a statement.
Murphy said he couldn’t divulge many details around the company’s product road map, but he expects Cornelis can get a next-generation high-speed interconnect to market within the next two to three years that will offer a level of bandwidth consistent with industry expectations.
“These things take between two and three years, and that’s only if you have a fantastic starting point,” he said of the intellectual property acquired from Intel. “Frankly, it would be impossible to try to tackle this without that.”
He added that Cornelis is working on a variety of technologies to accelerate different aspects of interconnects, including congestion control and dynamic adaptive routing. The company is also looking at how it can accelerate AI workloads from an interconnect perspective, which Murphy and his team believe is a “huge opportunity” in the future.
“What I see happening is that the techniques that have been long in play in high-performance computing are starting to become more and more adopted in artificial intelligence,” he said. “How do we scale this thing out? Well, guess what? We solved all those problems starting 20 years ago in high-performance computing. And so we just see this market coming towards us as opposed to us chasing that market, which is really exciting.”
One executive at a system builder company said his company was hoping Intel would find a way to keep Omni-Path alive because he has big customers that would like to continue using the technology, so the fact that Cornelis will continue what Intel started is good news.
“We were incredibly nervous about how to do existing support for our current install base,” he said.
With Cornelis being an independent company, it opens up the kind of technology partnerships the company can form to grow Omni-Path support. For instance, Intel’s rival in the x86 CPU space, AMD, would be able to work with Cornelis on integration for future Omni-Path products.
“Now that OPA is separate, it could work with Intel and it could go work with AMD platforms,” he said.
But even if Omni-Path is now a neutral technology, the executive questions whether Nvidia, whose GPUs are becoming more prevalent in high-performance computing, would be open to integration since it owns the only direct competitor to Omni-Path, the Mellanox InfiniBand interconnect. And with Nvidia’s plan to acquire Arm, the executive said, the potential for Omni-Path integration with Arm CPUs is up in the air, even if Nvidia has stated it plans to keep Arm technology open to all companies.
“We don’t know how it’s going to work with Arm when it’s under Nvidia. Now I know my understanding is Nvidia is going to continue licensing Arm to others,” he said, “So there’s a possibility that other Arm manufacturers could still work with OPA to get their stuff to work.”
However, the executive said, Cornelis still has a lot of potential as a stand-alone company and could see the sort of growth that could give Intel second thoughts on divesting.
“I just feel like [Intel CEO Bob Swan] really underestimated the potential of growth,” he said. “That could be a big mistake, and it could end up becoming a multibillion-dollar company.”