Next-generation NVIDIA Vera Rubin platform series
NVIDIA Vera Rubin NVL72 Platform
In consideration of the fact that next year’s key products are confidential, specifications cannot be provided at this time. Please refer to last year’s content as a reference:
NVIDIA Vera Rubin NVL72 Platform
Hon Hai showcased the next-generation NVIDIA Vera Rubin NVL72 AI computing platform, featuring a 48RU fully liquid-cooled rack design developed to meet the high-density computing requirements of next-generation large-scale AI Factories. Each rack integrates 72 NVIDIA Rubin GPUs and 36 NVIDIA Vera CPUs. Leveraging sixth-generation NVIDIA NVLink high-speed interconnect technology, the platform delivers a high-bandwidth, low-latency scale-up computing architecture to support large-scale AI model training and inference workloads.
The front of the system features 18 Vera Rubin NVL72 Compute Trays, 9 sixth-generation NVIDIA NVLink Switch Trays, and 2 TOR switches, along with four 3RU, 110kW Power Shelves. The rear integrates a Rack Manifold, 2,900A Busbar, and Cable Cartridge, enhancing power delivery, thermal management, and cable management capabilities in high-power-density environments.
The Vera Rubin NVL72 Compute Tray adopts a 1RU fully liquid-cooled and cableless design. Each tray houses two Vera Rubin Superchips, comprising a total of four Rubin GPUs and two Vera CPUs, and supports four E1.S NVMe drives. For networking, each tray can be equipped with one NVIDIA BlueField-4 DPU, providing up to 800 Gb/s of bandwidth, as well as up to four ConnectX-9 SuperNICs, with each GPU supporting up to 1.6 Tb/s of network throughput.
The platform integrates computing, networking, power delivery, and liquid cooling into a highly integrated system, demonstrating Hon Hai’s capabilities in next-generation AI server rack-level system design, liquid cooling, high-speed interconnects, and production integration. It provides large-scale AI Factories with a high-density, high-performance, and scalable infrastructure solution.