Built Exclusively for Agentic AI Workloads

In an exclusive disclosure, Nvidia has laid out the full specifications of its next-generation Vera CPU, a server processor that caters to the increasing needs of agentic AI. While most CPUs have been built for general purposes of computation, Vera is tailored exclusively to the specific requirements of AI systems that run on code, make use of tools, access data, and manage complex tasks.

At his GTC Taipei 2026 keynote speech, Jensen Huang, Nvidia’s CEO, revealed that the Vera can fetch and process up to 10 instructions per clock cycle. With this feature, Nvidia claims that Vera boasts outstanding instruction-per-clock (IPC) efficiency.

Next-Level Architecture and Ultrafast Speeds

Vera comes equipped with 88 proprietary Olympus cores that use Nvidia’s home-grown CPU architecture. These cores support dual threads simultaneously, delivering a combined total of 176 processing threads. The CPU boasts 2MB L2 cache per core and 164MB shared L3 cache to enhance data retrieval speeds.

According to Nvidia, Vera can deliver up to 1.8 times the speed of top x86-based CPUs during agentic AI benchmarking. It ensures efficient performance of latency-sensitive workloads while staying within a flexible thermal design power of 250W to 450W.

Massive Bandwidth, Wide-Scale Implementation Strategy

For increased efficiency, Nvidia has opted for a single compute die over a multi-chiplet architecture. It minimizes the communication delay while providing an advantage when moving data throughout the processor. The Vera chip comes with up to 3.4TB/s of bisection bandwidth due to Nvidia’s second-generation Scalable Coherency Fabric.
In addition, it is the first-ever Nvidia CPU for servers that supports LPDDR5X memory. The memory bandwidth offered by this chip can reach up to 1.2TB/s, along with a memory capacity of 1.5TB. The Vera processor will form the base of Nvidia’s upcoming Vera Rubin AI processor and is set to be used in standalone servers, liquid-cooled racks, and hyperscale AI.

FAQs

1. What is Nvidia Vera?
It is a data center CPU designed specifically for agentic AI and AI infrastructure.

2. How many cores does Vera have?
Vera features 88 Olympus cores and supports 176 threads.

3. What is Vera’s key advantage?
Its combination of high IPC, low latency, and exceptional memory bandwidth for AI workloads.

Conclusion

Vera represents Nvidia’s efforts to go beyond GPUs in developing their AI ecosystem. The processor combines unique design features, high bandwidth, and dedicated AI features in one system. It can be considered the cornerstone of future AI factories.