NVIDIA Vera: The First CPU Architecture Designed for AI Agents Ships at Scale
NVIDIA has begun shipping 'Vera,' its first CPU specifically designed to power autonomous AI agents. The chip architecture optimizes for the high-bandwidth needs of trillion-parameter workloads.
The era of the general-purpose CPU is giving way to the era of the 'Agentic CPU.' NVIDIA has officially started shipping 'Vera,' its first silicon architecture built from the ground up to support AI agents. As AI moves from simple chatbots to autonomous entities that can plan and execute complex tasks, the underlying hardware must evolve to handle the massive data throughput and low-latency requirements of trillion-parameter models.
Vera is designed to work in tandem with NVIDIA’s GPUs, but it takes over the critical 'reasoning' and 'orchestration' tasks that traditional CPUs struggle with under AI workloads. By integrating custom high-bandwidth memory (NVHBM) and the latest NVLink Fusion technology, Vera eliminates the bottlenecks that typically occur when moving data between the processor and the accelerator. This allows AI agents to maintain a 'live' state, processing environmental inputs and internal logic simultaneously without delay.
The deployment of Vera at scale marks a turning point for the semiconductor industry. We are seeing a shift toward heterogenous compute environments where every piece of silicon is specialized for a stage of the AI lifecycle. For developers of Physical AI and autonomous systems, Vera provides the 'brain' capable of managing the immense complexity of real-world interaction.
Source: NVIDIA Blog
