putilov_denis - stock.adobe.com
Cisco, Nvidia expand AI infrastructure partnership
The broader Nvidia-Cisco partnership includes Cisco building systems that combine its OS with Nvidia's Ethernet technologies for optimized AI performance.
Nvidia and Cisco expanded their partnership to develop Ethernet-based AI systems for enterprise data centers.
The broader partnership, unveiled Tuesday, will include Cisco building systems that combine Nvidia's Ethernet switch application-specific integrated circuits (ASICs) with Cisco OS software. The Nvidia Spectrum ASICs provide Ethernet networking for AI applications and support port configurations with speeds up to 400 GbE.
Nvidia also agreed to use Cisco Silicon One as the only third-partner processor within the Nvidia Spectrum-X Ethernet platform for AI workloads running in hyperscale clouds. The Nvidia Spectrum-4 ASIC in the networking platform is optimized for GPU cluster interconnects. Silicon One, which competes with networking chips from Broadcom, has a more versatile architecture designed for a broader range of use cases, including routing, switching, AI and machine learning.
The partnership aims to provide enterprises with AI infrastructure that can fit into their Ethernet environments without major changes in the management tools and processes used to operate networks, Cisco said. The infrastructure's architecture will span front-end networks designed to be accessible to a company's customers, partners and employees, and the back-end networks that manage internal processes, including data storage, computing and communication between application components.
Chuck Robbins
"Enterprises are under immense pressure to deploy AI quickly and effectively, and many leaders struggle to justify the investment while balancing the risks," Cisco Chairman and CEO Chuck Robbins said in a statement. "Together, Cisco and Nvidia are partnering to remove barriers for customers and ensure they can optimize their infrastructure investments to unlock the power of AI."
Nevertheless, the partnership raises questions neither company has addressed, Gartner analyst Arun Chandrasekaran said. Nvidia is heavily invested in InfiniBand and Ethernet, two networking technologies with different strengths. InfiniBand excels in high-performance computing (HPC), while Ethernet is more versatile and widely adopted for general-purpose networking.
Cisco is a founding member of the Ultra Ethernet Consortium, an industry group dedicated to developing an advanced Ethernet-based communication stack for HPC. It's unclear how Nvidia will position Ethernet and InfiniBand in the market.
"How they minimize this conflict and positioning will be interesting to watch," Chandrasekaran said.
Also, Nvidia has partnerships with Dell and Hewlett Packard Enterprise, which are both competitors of Cisco in the server market and the broader IT infrastructure space. Dell and HPE sell AI servers using Nvidia technologies.
"How much strategic air time will Cisco get within the Nvidia portfolio?" Chandrasekaran said. "There are lots and lots of partnerships that are announced in the industry, but which of them are true partnerships, where you're committing resources, and there is a very aligned go-to-market motion in place."
Upcoming technology
Cisco plans to develop data center switches that enable enterprises to standardize on the Nvidia Spectrum-X networking platform. The two companies also plan to collaborate on reference architectures based on Spectrum-X with several Cisco technologies, including Silicon One, Nexus switches, UCS Compute servers and Nexus Hyperfabric, which is cloud-managed infrastructure for generative AI models.
Introduced last June, Cisco developed Hyperfabric with Nvidia. The prebuilt product, which Cisco plans to release in May, includes Nvidia GPUs and BlueField-3 data processing units, Cisco Ethernet switching and a Vast data store.
In mid-year, Cisco plans to release Silicon One switches compatible with Spectrum-X and Nvidia reference architectures. The timeline for Cisco switches using Nvidia Spectrum ASICs was not released.
Antone Gonsalves is an editor at large for Informa TechTarget, reporting on industry trends critical to enterprise tech buyers. He has worked in tech journalism for 25 years and is based in San Francisco. Have a news tip? Please drop him an email.
