Nvidia Launches 102.4T Spectrum-6 Switch Designed for Vera Rubin Platform

Nvidia recently launched the next-generation Ethernet switch Spectrum-6 for gigawatt-scale AI factories, doubling network bandwidth compared with the previous generation. The switch has a capacity of 102.4T and is designed for the Vera Rubin platform, forming the core of the new Spectrum-X Ethernet platform. CoreWeave, Microsoft, Nebius, SpaceX AI and Tesla are among the first AI infrastructure builders to adopt the product.

Compared with traditional general-purpose Ethernet, the Spectrum-X Ethernet platform delivers 1.6 times the AI network performance and maintains up to 95% network efficiency at deployments of over 100,000 GPUs. The hardware-accelerated Spectrum-X multi-plane topology reduces the number of switches required in a data center by 40%, while silicon photonics technology brings a fivefold improvement in energy efficiency and a tenfold increase in mean time between failures.

Spectrum-6 is the network layer component of Nvidia's Vera Rubin overall computing platform, which integrates the Vera CPU, Rubin GPU, NVLink 6 switch, ConnectX-9 SuperNIC, BlueField-4 DPU and Spectrum-6 Ethernet switch. The Spectrum-6 switch chip combined with the ConnectX-9 SuperNIC forms the hardware foundation of the new generation Spectrum-X Ethernet. On the software side, the platform supports intelligent traffic load balancing and fast network fault bypass recovery. In terms of form factor and thermal design, Spectrum-6 supports both pluggable and co-packaged optics specifications, and supports liquid cooling solutions that can be integrated into an end-to-end liquid cooling deployment in AI factories.