IT House reported on March 17 that today (March 17) at the 2026 GTC conference in San Jose, California, USA, NVIDIA announced the detailed parameters of the new 88-core Vera data center CPU, claiming to be the world's first processor customized for intelligent AI and reinforcement learning.
When processing large-scale data, AI training and inference, the chip operates twice as efficiently as traditional rack-level CPUs and is 50% faster. NVIDIA founder and CEO Jensen Huang emphasized that the CPU is no longer just an auxiliary AI model, but has become the core driving force. Vera will help AI systems achieve faster "thinking" and wider expansion.
In terms of core architecture, the Vera data center CPU is equipped with 88 Olympus cores custom-designed by NVIDIA. In order to meet the stringent requirements of multi-tenant AI factory concurrent tasks, each core supports spatial multi-threading technology and can stably run two tasks at the same time.
In addition, Vera uses a second-generation low-power memory subsystem. The system is built on LPDDR5X memory and has a bandwidth of up to 1.2 TB/s. Compared with general-purpose CPUs, it significantly reduces power consumption by half while doubling bandwidth.
In order to meet the extreme expansion needs of data centers, NVIDIA simultaneously launched the Vera CPU rack based on the MGX modular architecture. The rack integrates 256 liquid-cooled Vera CPUs, which can maintain more than 22,500 independent concurrent computing environments running at full speed, more than 45,000 independent threads and 400TB of ultra-large memory. It not only achieves a 6-fold increase in CPU throughput, but also directly doubles the performance of intelligent AI workloads.
At the data transmission level, Vera is deeply paired with the GPU through NVLink-C2C interconnection technology, providing consistent bandwidth of up to 1.8 TB/s, which is 7 times that of PCIe 6.0.
Vera CPU has now entered full mass production and is expected to begin batch delivery to core customers such as Meta and Oracle in the second half of this year. IT Home attaches relevant screenshots as follows:
NVIDIA GTC 2026 Conference Topics
