Oracle has integrated Nvidia's standalone Vera CPUs into its multi-gigawatt data center infrastructure. This marks the first time Nvidia has deployed its own CPU as a standalone offering within a major hyperscaler environment, independent of GPU-focused clusters.
Nvidia's Vera CPU utilizes LPDDR5X-9600 memory and a distributed mesh architecture for caching. Testing shows the chip achieves a memory bandwidth of 1.2 TB/s, which exceeds the performance of dual Intel Xeon 6980P systems equipped with 8,800 MT/s MRDIMMs.
The chip is designed to support agentic AI workloads, including Python-based frameworks like CrewAI and LangGraph. Performance testing indicates that the architecture provides advantages in data-intensive tasks, such as large database queries and SQLite operations, due to the second-generation Scalable Coherency Fabric.
While integrating Vera, Oracle continues to expand its OCI Superclusters. The company plans to offer Nvidia HGX NVL8 configurations, which pair eight Rubin GPUs with dual 64-core Intel Xeon 6 hosts featuring 4TB of DDR5 memory.
