NVIDIA Vera CPU Is ‘Packing a Heavy-Hitting Punch’ Against Competition
Original reporting by NVIDIA Blog
The advent of agentic AI is rapidly reshaping the landscape of data center hardware, demanding a new class of CPU engineered for sustained high performance, fast cores, and immense memory bandwidth. Today, initial benchmark results from technology review site Phoronix reveal that NVIDIA's new Vera CPU is not just meeting this challenge, but fundamentally redefining expectations.
Designed from the ground up for the sequential and branch-heavy workloads characteristic of agentic AI, Vera integrates 88 custom NVIDIA Olympus cores, supported by a groundbreaking 1.2 TB/s of memory bandwidth. Phoronix's comprehensive testing shows Vera delivering aggressive throughput within an optimized power envelope, particularly excelling in CPU-intensive tasks like code compilation, database management, and data processing — precisely the operations central to AI factories.
Phoronix Confirms Leadership
Michael Larabel, founder of Phoronix, declared Vera "the most formidable competition to Intel and AMD x86_64 processors ever realized." The CPU demonstrated an "incredible advantage" in memory performance, leveraging a second-generation LPDDR5X subsystem to deliver up to double the peak memory bandwidth of traditional CPUs at a fraction of the power. Furthermore, Vera exhibited a remarkable 1.6x generational increase over NVIDIA's prior Grace CPU and outpaced a leading 128-core x86 processor by 1.5x in overall performance. With Vera now shipping to leading AI companies and cloud providers, its arrival signals a profound shift in the CPU competitive landscape, poised to power the next generation of intelligent systems.
NVIDIA's Vera CPU, powered by the custom Olympus cores, unequivocally demonstrates its readiness to power the burgeoning era of agentic AI. As benchmark results from Phoronix reveal, Vera is a CPU engineered from the ground up to meet the specific and demanding requirements of these workloads: high core utilization, immense and sustained memory bandwidth, and superior power efficiency. Delivering significant generational leaps over its predecessor and decisively outperforming leading x86 competitors in key agentic tasks like code compilation and database management, Vera establishes a new benchmark for data center CPUs designed specifically for modern AI factories. Its innovative LPDDR5X memory subsystem, in particular, showcases an "incredible advantage" in performance per watt, ensuring optimal operational efficiency for scalable AI infrastructure.
A New AI Foundation
The emergence of Vera signifies more than just a new CPU; it marks a pivotal moment for the AI industry and the broader computing landscape. NVIDIA’s aggressive and successful entry into the general-purpose data center CPU market fundamentally reshapes the competitive landscape, challenging long-standing incumbents with a purpose-built architecture optimized for the unique demands of AI. This robust performance validation of Armv9.2-compatible designs provides a powerful and efficient foundation for the scalable deployment of increasingly complex AI agents, potentially accelerating their development and adoption across diverse industries. Vera's remarkable efficiency and throughput promise to lower the operational barriers for building and scaling AI factories, making advanced agentic capabilities more accessible and economically viable, thereby driving the next wave of AI innovation and practical applications into reality sooner.