Microsoft has received the first production Vera Rubin systems from Nvidia, confirming its status as one of the earliest adopters of this next-generation AI architecture. These NVL72 systems combine 88-core Vera CPUs with Rubin GPUs featuring up to 288 GB of HBM4 memory each, along with liquid cooling solutions and NVLink networking. Nvidia claims the platform delivers up to 5x faster inference and 3.5x improved training capabilities compared to the previous Blackwell generation. Deployments are planned at new superfactory sites in Wisconsin and Atlanta, where CoreWeave has already reported a 10x increase in AI throughput per megawatt.
Source: Read the original article
