NVIDIA Vera CPU Launched for Agentic AI Workloads
NVIDIA has introduced the Vera CPU, a processor built specifically for agentic AI and reinforcement learning workloads. According to NVIDIA, it delivers results with 50% faster performance and twice the efficiency compared to traditional rack-scale CPUs.
Technical Specifications
The Vera CPU features 88 custom NVIDIA-designed Olympus cores, each capable of running two tasks using NVIDIA Spatial Multithreading. It includes a high-bandwidth memory subsystem built on LPDDR5X memory and uses the second-generation NVIDIA Scalable Coherency Fabric for faster agentic responses under high utilization conditions.
System Configurations
- New Vera CPU rack integrates 256 liquid-cooled Vera CPUs
- Sustains more than 22,500 concurrent CPU environments running independently at full performance
- Built using NVIDIA MGX modular reference architecture
- Part of NVIDIA Vera Rubin NVL72 platform with NVIDIA GPUs connected via NVIDIA NVLink-C2C interconnect
- Provides 1.8 TB/s of coherent bandwidth (7x PCIe Gen 6 bandwidth)
- Also serves as host CPU for NVIDIA HGX Rubin NVL8 systems
- Systems integrate NVIDIA ConnectX SuperNIC cards and NVIDIA BlueField-4 DPUs
Adoption and Partners
Customers collaborating with NVIDIA to deploy Vera CPU include Alibaba, ByteDance, Meta, Oracle Cloud Infrastructure, CoreWeave, Lambda, Nebius, and Nscale. Manufacturing partners include Dell Technologies, HPE, Lenovo, Supermicro, ASUS, Compal, Foxconn, GIGABYTE, Pegatron, Quanta Cloud Technology (QCT), Wistron, and Wiwynn.
Target Workloads
Vera systems are designed for reinforcement learning, agentic inference, data processing, orchestration, storage management, cloud applications, and high-performance computing. Systems partners provide both dual and single-socket CPU server configurations.
According to Jensen Huang, NVIDIA's CEO, "The CPU is no longer simply supporting the model; it's driving it. With breakthrough performance and energy efficiency, Vera unlocks AI systems that think faster and scale further."
📖 Read the full source: HN AI Agents
👀 See Also
Discovered Materials: AI Agents Discover 500+ New Materials, But Only 1 Has a Plausible Synthesis Route
Discovered Materials (YC P26) used frontier LLMs to computationally discover 500+ new semiconductor materials. Only 1 has a plausible synthesis route, highlighting the lab-to-fab gap.
Adaptive Inference Routing Proposal for AI Query Efficiency
A proposal submitted to Anthropic in April 2026 outlines a five-step system for routing queries to appropriate AI models based on complexity scoring, using simple signals like character count and sentence count before any model inference occurs.
Agora-1: Open Source Multi-Agent World Model for Real-Time Shared Simulation
Odyssey releases Agora-1, a world model that enables up to four agents (human or AI) to share a real-time generated simulation, using GoldenEye as the test environment.
What's missing in the 'agentic' story: a well-defined user agent role
Mark Nottingham argues that current AI agents lack a clear user agent role, creating a trust gap between what users expect and what agents actually do.