NVIDIA Vera

Specifications

Vera CPU — NVIDIA's First Arm Server Processor

Vera is NVIDIA's first server CPU built on custom Arm-compatible cores, serving as the CPU half of the Vera Rubin platform. The 88-core, 176-thread design introduces spatial multi-threading and pairs 1.5 TB of system LPDDR5X memory with 1.2 TB/s of memory bandwidth and a 1.8 TB/s NVLink-C2C coherent chip-to-chip interconnect to GPUs, alongside hardware confidential computing.

NVIDIA positions Vera as "the best CPU for agentic AI workloads" — the CPU that performs tool calling, SQL queries, and code compilation that keep GPUs and the wider data center fully utilized. Against x86, NVIDIA claims 3× more memory bandwidth per core, 2× energy efficiency, and 50% more single-threaded performance.

Vera ETL256 Rack

The Vera ETL256 is NVIDIA's first CPU-only rack, packaging 256 Vera CPUs with up to 400 TB of LPDDR5X, 300 TB/s of memory bandwidth, and 64 BlueField-4 DPUs, supporting 22,500+ concurrent CPU environments. Liquid-cooled and delivering 2× performance versus the previous-generation Grace CPU rack across scripting, text conversion, code compilation, and data/graph analytics, it is the dense agentic-AI host tier of NVIDIA's rack portfolio.

Application & Availability

Vera ships in 2H 2026 through cloud providers Lambda, Oracle Cloud, and Nebius, and through OEMs Dell Technologies, HPE, Cisco, Lenovo, and Supermicro. It anchors NVIDIA's Vera Rubin NVL72 (36 Vera CPUs + 72 Rubin GPUs) and the standalone Vera ETL256 CPU rack.

QS Compute supplies NVIDIA Vera processors and Vera-based server platforms for AI host-node, agentic-AI, and Arm-native cloud and HPC deployments.