NVIDIA has begun hand-delivering the first custom Vera CPU systems to major artificial intelligence firms. The initial recipients include Anthropic, OpenAI, and Oracle Cloud Infrastructure. Elon Musk accepted a delivery on behalf of his companies, which include SpaceXAI. This distribution marks the start of a global rollout in the coming months.
Major AI firms receive custom chips designed for agentic workloads
The Vera CPU features 88 custom Olympus cores designed for high-performance computing. It delivers 1.2 TB/s of memory bandwidth. The chip is the first data center CPU to utilize LPDDR5X memory. NVIDIA claims the architecture provides 50% faster per-core performance under full load compared to previous generations.
Spec comparison
| Spec | Vera |
|---|---|
| Core Count | 88 custom Olympus cores |
| Memory Bandwidth | 1.2 TB/s |
| Memory Type | LPDDR5X |
| Performance Improvement | 50% faster per-core performance under full load |
NVIDIA positions the Vera CPU as a processor purpose-built for Agentic AI tasks. These tasks include long-context retrieval and Python code generation. Ian Buck, NVIDIA Vice President of Hyperscale and High-Performance Computing, stated that Agentic AI is creating a new CPU moment in the AI factory. He noted that demand for CPUs is skyrocketing as models move from answering questions to taking action.
The Vera and Rubin platforms are currently entering full production phases. The company describes Vera as a new class of CPU designed to handle concurrent, real-time tasks. Traditional core-density focused designs were not built to prioritize the pressure these workloads place on processors. The Vera CPU aims to keep this work moving at scale for enterprise clients.
Source: TweakTown
Discussion
0 comments