---
格式版本: 2
标题: "Overview — NVIDIA NVL72 AI Factory"
原文链接: "https://docs.nvidia.com/enterprise-reference-architectures/nvl72-ai-factory/latest/overview.html"
发布日期: "未标注"
发现时间: "2026-06-01T16:19:30+08:00"
入库时间: "2026-06-01T14:37:13.994Z"
来源平台: "SerpApi Google"
搜索渠道: "serpapi"
搜索词: "NVL72"
匹配关键词:
  - "NVL72"
  - "NVIDIA"
  - "GPU"
  - "Nvlink"
相关厂家:
  - "NVIDIA"
相关专家:
  []
内容类型: "网页"
抓取工具: "XCrawl Scrape"
清洗工具: "XCrawl Markdown + LLM 正文裁剪"
原始附件:
  []
AI优质: "是"
AI打分: 93
AI分档: "高置信优质"
AI质检状态: "通过"
AI打分理由: "官方NVIDIA企业参考架构文档，直接聚焦GB300 NVL72机柜级AI系统。详细涵盖72卡GPU+36CPU拓扑、第五代NVLink互连、液冷设计及OEM交付与软件订阅模式，技术架构与商业部署信号明确，来源权威且内容完整，属高价值一手…"
AI质检模型: "qwen3.6-plus"
AI质检时间: "2026-06-02T23:16:49+08:00"
AI主题相关性: 20
AI来源权威性: 15
AI新颖性: 18
AI技术细节: 17
AI商业部署信号: 14
AI完整性: 9
采集批次: "2026年6月1日16点18分53秒"
采集批次ID: "20260601-161853-741"
去重键: "https://docs.nvidia.com/enterprise-reference-architectures/nvl72-ai-factory/latest/overview.html"
---

# Overview

The NVIDIA Enterprise RA using 2-4-5-800 (dual plane) node architecture with NVIDIA GB300 NVL72 and NVIDIA Spectrum-X Networking offers a fully integrated, rack-scale solution optimized for the most demanding AI workloads. It documents a modular architecture based on NVIDIA-Certified NVL72 GB300 systems, each with 72 NVIDIA Blackwell Ultra GPUs and 36 NVIDIA® Grace CPUs. With a liquid-cooled design, each Scalable Unit (a single NVL72 rack) integrates 18 compute trays connected via the fifth generation of NVLink. While each tray (single server or node) can still operate independently as needed, the NVLink interconnect enables GPUs to be dynamically combined, allowing the system to function as a single multi-GPU unit of compute for larger workloads. The design minimizes system bottlenecks through tightly coupled configuration to provide the best performance and application scalability. A fully tested system scales up to 8 SUs (Scalable Units). Larger clusters can be built based on customer requirements. The rack solution is delivered as a pre-configured system available through the fulfillment of OEM along with Hardware support. Software support from NVIDIA is based on a per GPU paid subscription of NVIDIA AI Enterprise. While this Enterprise RA document may make the solution look simple, it is highly recommended to utilize NVIDIA Mission Control for operation of the NVIDIA GB300 NVL72. The NVIDIA GB300 NVL72 and NVIDIA Spectrum-X Platform RA have been developed to support enterprise-grade AI deployments, with the following use cases:

* **Inference** on today’s largest AI models, in real-time
* **Training** and finetuning of trillion-parameter language models
* **Data analytics** that’s ultra-fast, at massive scale
* **High-performance computing** for single and mixed precision

For all use cases, this architecture is ideal for multi-user, single tenant workloads. Specifically, the logical design and software is streamlined for deployment and maintenance ease by tailoring the configuration to one where users are all part of the same enterprise, and accounting and access control can be consolidated. This RA is architected to support deployment of Kubernetes, Slurm, and associated applications and tools for non-virtualized workloads.