88 cores + 1.2TB/s bandwidth: NVIDIA's first-generation intelligent agent AI processor Vera

K-LinePoet · 2026-03-24T08:39:36+00:00

Nvidia unveiled its 88-core Vera data center CPU at the 2026 GTC conference, specifically designed for AI and reinforcement learning with processing efficiency that is twice that of traditional CPUs and a 50% speed improvement. Vera supports multi-threading technology and memory bandwidth of up to 1.2 TB/s, achieving extremely high computing performance when combined with the MGX modular architecture. Deliveries to customers like Meta are expected in the second half of the year.

K-LinePoet

2026-03-24 08:39:36

Abstract generation in progress

IT House, March 17 — Today (March 17), at the 2026 GTC conference held in San Jose, California, NVIDIA announced detailed specifications for its new 88-core Vera data center CPU, claiming it is the world’s first processor specifically designed for AI agents and reinforcement learning.

When processing large-scale data, AI training, and inference, this chip operates at twice the efficiency of traditional rack-level CPUs, with a 50% increase in speed. NVIDIA founder and CEO Jensen Huang emphasized that CPUs are no longer just auxiliary components for AI models but have become a core driving force. Vera will help AI systems achieve faster “thinking” and greater scalability.

In terms of core architecture, the Vera data center CPU features 88 NVIDIA-customized Olympus cores. To meet the demanding needs of multi-tenant AI factories running concurrent tasks, each core supports spatial multithreading, capable of simultaneously running two tasks reliably.

Additionally, Vera uses the second-generation low-power memory subsystem based on LPDDR5X memory, with a bandwidth of up to 1.2 TB/s. Compared to general-purpose CPUs, it doubles bandwidth while significantly reducing power consumption by half.

To support extreme data center scalability, NVIDIA also introduced the Vera CPU rack based on the MGX modular architecture. This rack integrates 256 liquid-cooled Vera CPUs, capable of maintaining over 22,500 independent full-speed concurrent computing environments, with more than 45,000 independent threads and 400TB of massive memory. This not only achieves a sixfold increase in CPU throughput but also doubles the performance of AI agent workloads.

In terms of data transfer, Vera uses NVLink-C2C interconnect technology paired with GPUs to provide a consistent bandwidth of up to 1.8 TB/s, seven times that of PCIe 6.0.

The Vera CPU is now in full production, with mass deliveries expected to begin in the second half of this year to key customers such as Meta and Oracle. IT House has attached relevant screenshots below:

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.