Engineered for the Agentic AI Era
The XA P2N-E2 is the first server platform powered by NVIDIA Vera, the first CPU purpose-built for AI agents. Each Vera CPU integrates 88 custom NVIDIA Olympus cores with Spatial Multithreading, delivering 176 threads of agentic throughput.
- 3.4 TB/s core-to-core bandwidth
- 1.8 TB/s NVLink™-C2C support for CPU-to-CPU communication
- 50% lower memory power consumption
The NVIDIA Vera CPU handles the CPU-bound steps on every AI agent's critical path — sandboxed code execution, tool use, orchestration, and data processing — so GPUs stay dedicated to training and inference.
- AI Model / Agent (NVIDIA RTX PRO™ Series GPU) sends tokens, code, and queries
- Sandbox Execution (NVIDIA Vera CPU) returns evaluations, results, and data
- Result: less time agents spend waiting, more time spent acting
The XA P2N-E2 is powered by two NVIDIA Vera processors — the first CPU built for AI agents. Each Vera CPU is built on 88 custom NVIDIA Olympus cores with Spatial Multithreading for 176 threads of agentic throughput, engineered to handle the CPU-bound steps on every agent's critical path: sandboxed code execution, tool use, orchestration, and data processing. An LPDDR5X memory subsystem delivering up to 1.2TB/s of bandwidth keeps active cores fed with data, while second-generation NVIDIA NVLink™-C2C interconnect provides up to 1.8TB/s of coherent CPU-to-CPU communication. Together, this lets the XA P2N-E2 execute agent sandbox workloads at scale while GPUs stay focused on training and inference — reducing the time agents spend waiting and increasing the time they spend acting.
The XA P2N-E2 uses a fully air-cooled thermal design to keep dual 480W-per-socket NVIDIA Vera processors and dual-slot GPUs running efficiently — no liquid-cooling infrastructure required. An optimized front-to-back airflow path across the HPM (CPU), SOCAMM2 memory, and GPU zones delivers lower power consumption per token generated, giving enterprises a simpler, lower-TCO path to deploying agentic AI at scale.
Fully compliant with NVIDIA MGX™ architecture, the XA P2N-E2 is engineered for rapid, large-scale deployment across Agentic AI infrastructure. Its modular design lets IT teams service, upgrade, or scale CPU and GPU resources independently — shortening time-to-deployment and giving customers a validated path to scale agentic AI capacity as demand grows.
1
2
1
GPU Bay
Houses up to two dual-slot NVIDIA RTX PRO™ 6000 Blackwell Server Edition GPUs, independently serviceable for fast scaling.
2
Host Processor Module (HPM)
Houses dual NVIDIA Vera processors and 16 x SOCAMM2 memory modules, independently serviceable.