Nvidia Open-Sources Vera Rubin Rack Designs and Partners With Meta, Google, and Cisco on AI Networking

Nvidia contributed its Vera Rubin rack and compute tray designs to the Open Compute Project via the MGX architecture to standardize AI factory power, cooling, and networking layouts. Meta Platforms is adopting Nvidia's 102.4 terabit-per-second Spectrum-6 Ethernet switch to link directly with Vera Rubin infrastructure. Nvidia will integrate Google's Falcon Transport protocol into its Spectrum-X Ethernet lineup starting with future ConnectX-10 SuperNICs and expanding to BlueField silicon. Partnering with Cisco Systems, Nvidia is co-developing an Ultra-dense Small Form Factor Pluggable optical module delivering four times the bandwidth density of current 200G OSFP components and a claimed 1.3-times improvement in power efficiency. The Vera CPU platform reportedly delivers roughly double the performance of prior Grace silicon on select benchmarks, alongside a 1.8-times gain on Terminal-Bench container orchestration tasks, as system architects target liquid-cooled, Frore diamond coldplate systems and high-density fabrics. "The real bottleneck for the industry right now is how fast the world can manufacture and deploy AI factories." — Joe DeLaere