NVIDIA’s New Vera Processor Outperforms AMD, Intel Chips in Early Server Tests

Early performance results for NVIDIA’s Vera processor demonstrate a technical advantage over AMD-made x86 architecture chips and the Intel. The data was released by the specialized portal Phoronix, which carried out a battery of evaluations with the new component aimed at corporate infrastructures. The equipment uses the ARM architecture and has 88 cores from the Olympus line, custom developed by the manufacturer. The introduction of this hardware represents a change in the dynamics of the data center sector, historically dominated by traditional semiconductor manufacturers.

Durante direct evaluations, the NVIDIA component recorded a 63% higher performance than its direct predecessor, the Grace CPU, which operated with 72 cores. In comparison with the competition, the Vera was 10% faster than AMD’s EPYC 9575F processor, equipped with 64 cores of the Zen 5 architecture running at 5 GHz. The difference was even greater in relation to the Intel Xeon 6980P, with the ARM chip outperforming the 128-core model of the Granite Rapids architecture by 55%. The test used the Geomean metric to consolidate the results, where the NVIDIA product reached the mark of 26.40 points, against 23.79 for the AMD competitor.

Foco on artificial intelligence and advanced architecture

The development of Vera took place with a specific focus on processing agentic AI applications and inference tasks. The manufacturer promises to deliver performance 50% higher than current corporate market standards. The project also aims to double the energy efficiency of operations and quadruple the processing density per server rack, when compared to conventional x86 processors. Optimizing for artificial intelligence workloads has become a key growth vector for hardware companies in recent years.

Large-scale production of Vera CPUs has already started by NVIDIA. The first batches of racks equipped with the new technology have started to be delivered to large corporations in the technology sector, including CoreWeave, Meta, Oracle and Alibaba. Estes processors are part of the Extreme Co-Design ecosystem, a technical framework that powers the company’s Rubin platform. The strategy marks the company’s consolidation in a segment of independent processors, expanding its portfolio beyond traditional graphics processing units.

The demand for infrastructure capable of supporting complex language models requires components that can process large volumes of data in real time. The availability of Vera processors in standalone format, without the obligation of an exclusive link with the Rubin line racks, expands the integration possibilities for corporate customers. Companies already using the NVIDIA ecosystem for artificial intelligence workloads form the main target audience for this initial rollout phase.

Restrições in the dissemination of energy metrics

Apesar of the positive results in processing speed, some fundamental technical indicators were left out of the initial publications. The Phoronix portal reported that it did not receive authorization from the manufacturer to perform or publish performance tests per watt. Esta metric evaluates component energy efficiency and represents a decisive factor for data center operators, who deal with high electricity and cooling costs in their cloud processing facilities.

The hardware used during the test battery was a module in the pre-production phase. The absence of consumption data suggests that NVIDIA’s engineering team is still working on power management optimizations and fine-tuning before the product’s definitive commercial launch. Mesmo with the information restrictions on electrical consumption, the specialized publication classified the equipment as the most performant ARM processor for Linux servers ever evaluated by its laboratory technicians.

Leia Tambem

Leak from accessory manufacturers of the unreleased iPhone Ultra confirms horizontal fold and lateral Touch ID

Entrepreneurs adjust digital tactics to convert engagement into real sales in today’s market

Experts reveal effective methods for blocking spam calls on cell phones and landlines

Especificações technical new enterprise hardware

The internal architecture of the processor reflects design choices aimed at massive parallelism. The 88 Olympus cores operate at balanced frequencies to prevent overheating in high-density environments. Cache memory capacity has received significant increases compared to the previous generation, reducing access time to critical data during inference operations. The increase in core density per socket allows servers to process multiple simultaneous requests with lower network latency.

• 88 custom Olympus ARM cores for high performance.
• Suporte the low latency memory architecture for agentic AI payloads.
• Consumo estimated energy between 200 and 300W per processing socket.
• Native Compatibilidade with the Linux operating system software stack.
• High-density Empacotamento designed for multi-processor racks.
• Optimized memory transfer Taxa for large language models.

Compatibility with Linux-based systems makes it easier for infrastructure administrators to adopt the hardware, without having to rewrite fundamental operating codes. The integration of very high-speed memories directly into the processor package eliminates traditional communication bottlenecks between the CPU and external memory banks. Estas technical features position the component as a high-throughput solution for modern data processing centers and high-performance computing platforms.

Movimentação from competition in the server industry

NVIDIA’s advance in the server CPU market generates direct reactions from competing companies. AMD is working on the development of the EPYC Venice line, based on the new Zen 6 architecture. Este hardware is already in the mass production phase and is expected to reach the corporate market in the second half of 2026. Intel, in turn, is accelerating the schedule for its Diamond Rapids platform to try to recover the market share lost in the last financial quarters.

The competitive landscape also includes companies with a tradition in mobile devices that are looking for space in data centers. Qualcomm and Arm develop their own specific processor designs for the agentic AI segment. Supplier diversification offers infrastructure operators more architectural options to build their processing clusters, breaking the historical dependence on the x86 architecture that dictated the rules of the corporate market in recent decades.

Projeções NVIDIA’s internal financials indicate that the Vera line could generate revenue of approximately US$ 20 billion throughout the year 2026. Achieving this financial target would place the company in the position of the largest CPU supplier in the world in terms of revenue volume. The transition from a graphics card-focused manufacturer to a full-service processing solutions provider shifts the balance of power in the global enterprise semiconductor industry.