The Nvidia Vera Rubin NVL72 systems and Vera CPUs of the new Nvidia hardware generation are now accessible on Oracle Cloud Infrastructure. This news was shared at the Oracle AI World event. In addition, Oracle plans to include the Nvidia Vera Rubin systems in its “Superclusters” service, which will involve a large-scale deployment of multi-gigawatts.
In the future, the Rubin HGX NVL8 Systems will be incorporated, featuring eight Rubin GPUs interconnected via NVLink within an x86-based server setup. – Getty Images. OCI Superclusters consist of Nvidia NVL72 rack-scale systems, Oracle Acceleron networking with Nvidia Spectrum-X Ethernet networking that includes Nvidia Spectrum-6 switches and Nvidia BlueField-50 DPUs, along with gigawatt-class power and cooling capabilities to train and deploy the largest AI models.
Every rack links 72 GPUs via Nvidia NVLink scale-up networking, and OCI’s robust, multi-layered network extends across numerous GPUs. The incorporation of Vera Rubin into Superclusters complements Oracle’s existing Nvidia Grace Blackwell computing capabilities, which were employed in Abilene by OpenAI to train GPT-6 Astra.
OpenAI utilized an OCI Supercluster encompassing over 100,22027 Blackwell GPUs. Sreekanth Pothanis, OpenAI’s engineering leader for compute infrastructure, stated, “As our frontier models progress, we require infrastructure that can keep up.” Nvidia’s Vera Rubin on OCI provides enhanced computing performance and scalability, enabling accelerated model training and efficient handling of growing AI workloads.
The introduction of Vera CPUs focuses on agentic AI tasks, merging superior single-thread functionality with memory bandwidth to expedite the completion of CPU-demanding activities for agents. According to Mahesh Thiagarajan, Oracle Cloud Infrastructure’s president, “Oracle Cloud Infrastructure provides customers with the performance, security, and scalability required to progress in frontier AI,” “Through our partnership with Nvidia, we are enhancing our AI solutions in training, inference, and agentic AI.
By integrating Nvidia’s Vera Rubin GPUs and Vera CPUs with Oracle’s Acceleron networking and AI-optimized data centers, we aim to maintain the productivity, security, and scalability of GPUs for our customers as their requirements evolve.” Oracle jest bardzo skupiony na procesorach.
