NVIDIA has delivered its first Vera CPUs to major technology firms including Anthropic, OpenAI, SpaceX AI, and Oracle Cloud. The company announced the release on May 19, 2024. This chip marks a significant shift in the company's processor strategy for artificial intelligence workloads.
Major tech firms receive new processor designed for high-throughput inference tasks.
The Vera CPU features 88 NVIDIA-designed Olympus cores. It delivers 50% higher single-thread performance compared to the previous generation Grace CPU. The processor supports FP8 precision for direct AI inference and reinforcement learning tasks. It also provides 1.2 TB/s of memory bandwidth to handle high-throughput data movement.
Oracle Cloud Infrastructure plans to deploy hundreds of thousands of Vera CPUs starting in 2026. Karan Batta, head of product management at Oracle, stated that agentic AI requires sustained performance at scale. He noted that Vera's architecture is designed for high-throughput inference to power the next generation of enterprise AI.
Vera serves as the host processor for the upcoming Vera Rubin superchip platform. It connects to two Rubin GPUs via NVLink-C2C interconnect technology. The chip focuses on high-throughput inference, tool calling, and code generation rather than maximizing core count. NVIDIA claims it offers two times the energy efficiency of traditional infrastructure for scheduling and control tasks.
Discussion
0 comments
Log in to join the thread with a thoughtful take, question, or correction.