News
IndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value ChainIndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value Chain
EveryChip
Nvidia Vera CPU

Nvidia Vera CPU

Arm-based data centre CPU for agentic AI workloads

Nvidia Vera is a central processing unit designed for agentic artificial intelligence and reinforcement learning. It uses custom Olympus cores and LPDDR5X memory to accelerate code execution and data processing in AI factories.

Share on WhatsApp

  • Features 88 custom Olympus cores for single-threaded performance
  • Uses LPDDR5X memory with up to 1.2 TB/s bandwidth
  • Monolithic die design avoids cross-chiplet communication latency
  • Delivers 176 threads via NVIDIA Spatial Multithreading
  • Connects to GPUs with 1.8 TB/s NVLink-C2C bandwidth

What it is

The Nvidia Vera CPU is a processor built for the specific demands of agentic artificial intelligence and reinforcement learning. Unlike traditional server chips, it focuses on single-threaded performance to handle tasks such as code compilation, tool usage, and sandboxing. Nvidia states that this design helps keep graphics processing units fed with data while managing orchestration loops. The chip serves as both a host processor for accelerated systems and a standalone unit for AI factory workloads. It aims to improve the efficiency of token generation and reduce latency in complex AI reasoning tasks.

How it came about

Nvidia developed Vera to address bottlenecks in modern AI factories where central processors struggle with branch-heavy workloads. The company positions it as a new class of CPU architected for the era of agents. Early testing involved using the chip to speed up the design of next-generation Nvidia hardware. Ian Buck, an executive at Nvidia, delivered the first systems to major technology firms including Anthropic, OpenAI, and Oracle Cloud Infrastructure. This marked the transition of the platform from announcement to production deployment in top AI laboratories.

Inside the chip

The processor features 88 custom Olympus cores designed for latency-sensitive tasks. It employs Spatial Multithreading to create 176 threads by partitioning core resources. A monolithic compute die avoids the latency penalties of cross-chiplet communication. All cores connect via the second-generation Scalable Coherency Fabric, which provides high bisectional bandwidth. The memory subsystem uses LPDDR5X modules in SOCAMM form factors, offering high bandwidth per watt. NVLink-C2C connectivity allows coherent data sharing with Nvidia GPUs, creating a unified memory architecture for mixed workloads.

How it was made

Nvidia designed the Vera CPU with a monolithic compute die structure. This approach integrates all cores, cache, and input-output components on a single piece of silicon. The company claims this design ensures predictable latency and throughput by eliminating delays associated with multi-chiplet links. The system supports detachable and field-replaceable memory modules, allowing for serviceability and capacity upgrades. Specific details regarding the semiconductor manufacturing process node or the foundry partner are not provided in the available material.

Where it went

This processor targets hyperscale cloud environments, enterprise data centres, and high-performance computing clusters. It runs critical infrastructure tasks such as extract-transform-load operations, key-value cache management, and agent orchestration. The chip also powers standalone servers and storage platforms like the Vera BlueField-4 STX. Nvidia reports that it accelerates verification and simulation workloads for chip design. Initial adopters include AI research labs and cloud providers seeking to improve the responsiveness of software agents and reinforcement learning systems.

Versions and family

The Vera CPU is part of a broader roadmap that includes the upcoming Rosa CPU with Rigel cores. It is available in dense, liquid-cooled rack configurations as well as standard dual-socket and single-socket setups. The processor integrates with Vera Rubin NVL72 and HGX Vera Rubin NVL8 platforms. These systems pair the CPU with Nvidia Rubin graphics processors to maximise throughput. The architecture also extends to storage infrastructure through specialized variants designed for AI data handling.

Why it matters

Nvidia claims the Vera CPU delivers significantly faster performance for agentic sandboxes compared to leading x86 processors. The company states it achieves up to five times the bandwidth per watt for memory power. By reducing the time required for code execution and tool calls, the chip aims to increase the overall output of AI factories. It extends the Nvidia AI platform from graphics processing units to central processing units, providing a full-stack solution for safe and efficient agent deployment.

Problems and reception

Industry innovators such as Perplexity have adopted the platform for their AI infrastructure. Technical reviews highlight its strong single-threaded performance and memory results for agentic tasks. The chip has been described as packing a heavy-hitting punch against competition in benchmark tests. Its arrival at major AI labs signals a shift towards specialized central processors for agent-based workflows. Users note its ability to handle concurrent sandbox environments and complex evaluation loops effectively.

Questions people ask about Nvidia Vera CPU

What is the primary purpose of the Nvidia Vera CPU?

It is designed for agentic AI and reinforcement learning workloads. The chip handles tasks like code execution, tool use, and data orchestration to keep AI factories running efficiently.

How many cores does the Vera CPU have?

The processor contains 88 custom-designed Olympus cores. It uses Spatial Multithreading to support 176 threads by partitioning resources within each core for predictable throughput.

What type of memory does the Vera CPU use?

It uses high-bandwidth LPDDR5X memory in SOCAMM modules. This setup provides up to 1.2 terabytes per second of bandwidth and supports capacities up to 1.5 terabytes.

Is the Vera CPU based on x86 or Arm architecture?

The Vera CPU is built on the Arm architecture. It offers full Arm compatibility and is designed to extend the Nvidia AI platform from GPUs to central processing units.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Nvidia Vera CPU yet, so the facts on this page come from the sources listed above: nvidia.com/en-us/data-center/vera-cpu/, tomshardware.com/pc-components/cpus/nvidia-touts-vera-cpus-single-thre. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryChip wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the chip of the day, new pages on EveryChip, and the conferences and anniversaries of the week ahead.