NVIDIA and Oracle plan multi-gigawatt build-out of Vera Rubin AI systems
The Santa Clara chipmaker says its newest AI rack systems and Vera processors are now available on Oracle's cloud.
NVIDIA, the Santa Clara chipmaker, said Oracle plans to grow its largest computing clusters for artificial intelligence (AI) to "multi-gigawatt scale" using NVIDIA's newest systems, called Vera Rubin NVL72, according to an NVIDIA news release dated Oct. 27.
A gigawatt is a billion watts, roughly the output of a large power plant, so the plan describes data centers that use huge amounts of electricity.
Each Vera Rubin NVL72 rack links 72 graphics processing units (GPUs), the chips that do most AI training and answering. NVIDIA said those racks and its Vera central processing units (CPUs), the general-purpose chips that run a computer, are now generally available, meaning any customer can rent them, on Oracle Cloud Infrastructure (OCI), Oracle's service for renting computing power online.
The companies also plan a smaller system with eight Rubin GPUs. The release said OpenAI trained and ran a model called GPT-6 Astra on an Oracle cluster in Abilene with more than 100,000 earlier NVIDIA Blackwell GPUs. "We are bringing Vera into production on OCI to support these workloads at scale," said Uday Ruddarraju, OpenAI's chief technology officer of compute.