Each individual Vera Rubin NVL72 rack contains 72 Rubin GPUs, 36 Vera CPUs, and 20.7 TB of HBM4 memory. The systems are designed for high scale-up bandwidth, providing 216 TB/s via NVLink to facilitate faster AI model training and inference.

Supermicro began shipping these Nvidia-powered units on September 23 and is also offering a 16 rack Scalable Unit featuring 1,152 Rubin GPUs and 331 TB of HBM4. This larger configuration includes a direct liquid cooling stack intended for data center deployments ranging from 5 MW to gigawatt scale.