Cisco today announced an expanded partnership with NVIDIA to provide customers with the flexibility to choose network infrastructures to support increasingly intensive AI workloads and high-performance data exchange between and within data centers, as well as between clouds and between users.
According to Chuck Robbins, chairman and CEO of Cisco, the collaboration will “remove barriers for customers” and allow them to “optimize their infrastructure investments to unleash the potential of AI.” NVIDIA CEO Jensen Huang added that AI is advancing at the “speed of light,” noting the transformative potential of combining Cisco’s global reach with NVIDIA’s AI-optimized Ethernet solutions to help companies around the world build a modern AI infrastructure.
Image source: Cisco
As part of the expanded collaboration, the NVIDIA Spectrum-X Ethernet networking platform, powered by Cisco and NVIDIA ASICs, will serve as the foundation for many enterprise AI workloads. By leveraging a unified architecture that simplifies the integration and standardization of Cisco and NVIDIA technologies in enterprise environments, customers can optimize their infrastructure investments using existing management tools and processes across different types of networks.
The partnership between the two companies will enable their customers to benefit from current and future technology advancements in the NVIDIA Spectrum-X platform, such as adaptive routing, telemetry, congestion management, and low latency, while also giving them access to Cisco’s broader portfolio of networking, security, and digital resilience solutions, including the Splunk platform. As companies develop AI capabilities and train models on their own data, having a comprehensive security strategy is critical, the press release noted.
Image source: NVIDIA
If we put aside the beautiful marketing words, the new cooperation comes down to two main points. First, the NVIDIA Spectrum-X platform will now include solutions based on Cisco Silicon One ASIC and NVIDIA SuperNIC DPU, and only Cisco chips will be allowed to the platform. Cisco itself already has a similar solution based on AMD’s Pensando DPU.
Secondly, Cisco is porting its proprietary NX-OS network OS to the NVIDIA Spectrum ASIC, which will allow standardizing network infrastructures in data centers, organizing a single point of control for the entire network, and integrating NVIDIA solutions with the Nexus stack and other Cisco services and products, including Nexus Dashboard, Nexus Hyperfabric AI, UCS, etc. Cisco itself will release products based on the NVIDIA Spectrum ASIC.
Image source: Cisco
Cisco and NVIDIA’s collaboration extends to the joint development of reference architectures, including the NVIDIA Cloud Partner (NCP) and Enterprise Reference Architectures. Cisco and NVIDIA also plan to continue collaborating to address critical challenges such as congestion management and balancing, and other operational issues that enterprises face as they scale AI workloads.
Interestingly, NVIDIA eventually joined the Ultra Ethernet consortium, founded by Arista, Broadcom, Cisco, and others to create a modern interconnect for HCP and AI workloads based on Ethernet. In fact, the consortium is largely intended to create a working alternative to Infiniband technology, which is monopolized and promoted by NVIDIA.