NVIDIA Vera

CPU for the Age of AI.

Overview
Use Cases
Features
Vera Architecture
Get Started

Overview
Use Cases
Features
Vera Architecture
Get Started

Overview

Purpose-Built for Agentic AI

NVIDIA Vera is built for reinforcement learning (RL) and agentic AI, powering the code, tools, and data workflows that operate beyond the model. As the host CPU in accelerated systems, Vera pairs seamlessly with NVIDIA GPUs—directing data movement, managing memory, and orchestrating system control to keep AI pipelines running at full speed. With high-performance, energy-efficient cores and massive LPDDR5X memory bandwidth, Vera enables software environments to run up to 50% faster with twice the efficiency of traditional CPU infrastructure, unlocking agentic AI at scale.

NVIDIA Launches the Vera CPU, Purpose-Built for Agentic AI

The NVIDIA Vera CPU delivers the highest performance and energy efficiency for data processing, AI training, and agentic inference at scale.

Read the Press Release

Inside the NVIDIA Rubin Platform: Six New Chips, One AI Supercomputer

Built through extreme codesign, NVIDIA Vera Rubin treats the data center, not the chip, as the unit of compute, establishing a new foundation for producing intelligence efficiently, securely, and predictably at scale.

Read the Tech Blog

NVIDIA Vera CPU Rack

The NVIDIA Vera CPU Rack powers reinforcement learning and agentic AI at AI factory scale. Built on NVIDIA MGX™, it integrates up to 256 Vera CPUs to run over 22.5K concurrent environments.

Learn More

Use Cases

Next-Generation Data Centers

AI Factories

NVIDIA Vera delivers system-level efficiency to AI factories, serving both as the host CPU in NVIDIA Vera Rubin NVL72 and HGX™ Rubin NVL8 platforms. Vera feeds GPUs for large-scale AI and serves as the compute backbone for the tasks that keep the factory running, including extract, transform, and load (ETL); key-value (KV) cache management; and orchestration. With high single-threaded performance, massive memory bandwidth, and a single compute die design that avoids cross-chiplet latency, Vera delivers predictable performance while keeping GPUs fully utilized across accelerated AI and HPC systems.

Core Compute Platforms

For RL and agentic AI, NVIDIA Vera delivers leading per-core performance and massive memory bandwidth to power thousands of parallel software environments. Built for control-heavy, latency-sensitive workloads, Vera enables 50% faster evaluation cycles under full load. Vera also operates as a high-performance standalone CPU platform for hyperscale cloud, analytics, storage, enterprise, and HPC workloads—providing leading performance and energy efficiency across data-intensive and real-time applications. Available as a dense liquid-cooled Vera CPU rack or in standard dual- and single-socket configurations, Vera meets the needs of any data center.

Features

Explore the Technological Breakthroughs

Built for the demands of RL and agentic AI, the Vera CPU combines high-performance NVIDIA custom-designed Olympus cores, efficient LPDDR5X memory, low-latency NVIDIA Scalable Coherency Fabric (SCF), NVIDIA NVLink™ Chip-to-Chip (C2C) connectivity, support for full confidential computing, and full Arm® compatibility. Vera’s monolithic compute architecture keeps software environments responsive and data flowing, maximizing throughput, energy efficiency, and GPU utilization across AI, analytics, and HPC workloads.

NVIDIA-Designed Olympus Cores With Spatial Multithreading

The NVIDIA Vera CPU features 88 Olympus cores, delivering 2x the performance of its predecessor with industry-leading energy efficiency and full Armv9.2 compatibility, and is the first CPU to support FP8 precision. Designed for RL and agentic AI, the Olympus cores deliver the leadership single-thread performance required to run control-heavy software environments at scale. Each core supports NVIDIA Spatial Multithreading, a new type of multithreading that enables 176 total threads by physically partitioning each core’s resources rather than time slicing them, allowing the system to optimize for performance or density at runtime.

Energy-Efficient Memory Subsystem With More Bandwidth and Capacity

The NVIDIA Vera CPU delivers up to 1.2 terabytes per second (TB/s) of memory bandwidth - twice the bandwidth at half the power compared to traditional CPUs. This massive bandwidth keeps thousands of parallel software environments responsive, enabling faster reinforcement learning iterations, efficient KV-cache management, and data-intensive agentic workflows. With support for up to 1.5 TB of memory—3x the prior generation—Vera provides the capacity and efficiency required for next-generation AI factories and memory-intensive analytics and HPC workloads.

Maximized Performance With Second-Generation NVIDIA SCF

Modern CPU workloads demand fast and predictable data movement under full load. The second generation of NVIDIA SCF—a 3.4 TB/s bisection bandwidth, on-chip mesh with a unified cache—scales all 88 cores across a single compute die, delivering uniform, high-bandwidth access to compute and memory. By keeping compute tightly integrated and avoiding cross-chiplet communication, SCF maintains consistent latency and throughput when all cores are fully utilized—ensuring stable, predictable performance at AI-factory scale.

Seamless Data Sharing With Second-Generation NVIDIA NVLink-C2C

NVIDIA NVLink-C2C delivers 1.8 TB/s of coherent bandwidth, enabling seamless data sharing between processors. When paired with NVIDIA Rubin GPUs, it forms a unified memory system that allows the CPU and GPU to work together on complex AI and HPC workloads, large datasets, and KV-cache offload, while supporting secure, hardware-enforced isolation for sensitive data and code. NVLink-C2C also reduces data-transfer bottlenecks and simplifies optimization in dual-socket NVIDIA Vera CPU systems.

NVIDIA Vera Rubin NVL72

NVIDIA Vera Rubin NVL72 unifies leading-edge technologies from NVIDIA: 72 Rubin GPUs, 36 Vera CPUs, ConnectX®-9 SuperNICs, and BlueField®-4 DPUs. It scales up intelligence in a rack-scale platform with the NVLink 6 switch and scales out with NVIDIA Quantum-X800 InfiniBand and Spectrum-X™ Ethernet to power the AI industrial revolution.

Learn More

Get Started

Stay Up to Date on NVIDIA News

Stay Informed

Email Me When Available

Section

Section

First Name

Last Name

Business Email Address

Organization / University Name

Industry

Job Title

Location

Preferred Language

State/Province

enterpriseOptIns hidden field

Send me the latest enterprise news, announcements, and more from NVIDIA. I can unsubscribe at any time.

nvid hidden field

ncid hidden field

NVIDIA Privacy Policy

I agree to the collection and processing of the above information by NVIDIA <span class="corporation-txt hidden">Corporation </span>for the purposes of research and event organization, and I have read and agree to <a href="https://www.nvidia.com/en-us/about-nvidia/privacy-policy/?deeplink=visiting-our-website" target="_blank">NVIDIA Privacy Policy</a>.

I agree that the above information will be transferred to NVIDIA Corporation in the United States and stored in a manner consistent with <a href="https://www.nvidia.com/en-us/about-nvidia/privacy-policy/?deeplink=visiting-our-website" target="_blank">NVIDIA Privacy Policy</a> due to necessities for research, event organization and corresponding NVIDIA internal management and system operation need. You may contact us by sending an email to <a href="mailto:privacy@nvidia.com">privacy@nvidia.com</a> to resolve related problems.