Nvidia Integrates Cisco into Spectrum-X Reference Architecture

Nvidia has announced an expansion of its collaboration with Cisco, incorporating the networking specialist into its Spectrum-X reference architecture for artificial intelligence (AI) infrastructure.

This new partnership marks the first time Nvidia has included a third-party vendor in its Spectrum-X reference architecture, which is specifically designed for AI networking solutions. Cisco plans to adapt its NX-OS, a network operating system tailored for data centers, to function on Nvidia's hardware platforms. In the future, both companies aim to enable management and monitoring through the Nexus dashboard, enhancing the overall user experience.

Historically, Nvidia and Cisco had not developed a joint product. However, in July 2024, Cisco announced a significant partnership with Nvidia at its annual conference in Las Vegas, introducing the HyperFabric AI Cluster. This initiative focuses on creating plug-and-play AI infrastructures. Although the entire Nvidia software stack was listed in Cisco's global pricing catalog, this collaboration had yet to yield tangible products until now.

Previously, Spectrum-X was exclusively designed for Nvidia's own hardware and software solutions, leaving out other manufacturers. According to company sources, the integration of Cisco's Silicon One ASIC family with Nvidia's BlueField-3 SuperNICs into the Spectrum-X Ethernet network platform is a key development. Cisco plans to incorporate its Nexus 9000 series switches with 400G and 800G ports into the reference architecture, addressing the high bandwidth and low latency requirements essential for AI training clusters in backend networks using RDMA over Converged Ethernet (RoCE).

Proper synchronization of individual components is crucial for optimal performance, which the reference architecture is expected to facilitate. Cisco is also set to provide optical solutions for the 400G and 800G ports, further enhancing the network capabilities.

This integration aims to streamline operations for AI developers by offering a cohesive architecture for AI workloads, allowing them to concentrate on their core tasks without the complexities of system interoperability.

The HyperFabric AI Cluster is not yet available; however, this partnership reflects the broader industry trend among major companies in the UltraEthernet Consortium to move away from Infiniband technology towards Ethernet solutions. This shift opens the door for additional Ethernet-based product manufacturers catering to AI networks.

For Cisco, the collaboration with Nvidia aligns well with its AI strategy. However, the company has recently experienced a decline in market share against competitors like Arista in the high-speed switch segment critical for AI networks. It remains to be seen whether this new partnership will bolster Cisco's sales, especially as major AI companies like Meta have recently opted for Arista switches to build their Ethernet-based AI networks.