Reference Architectures
Build Enterprise AI Factories That Scale
Turn your data center into a high-performance AI factory with NVIDIA Reference Architectures.
Overview
The Building Blocks for AI Success
NVIDIA Reference Architectures (RAs) enable organizations to design, deploy, and scale high-performance AI factories using validated, repeatable infrastructure. These designs combine certified compute, high-speed east-west and north-south networking, observability tools, and infrastructure software to ensure scalable performance, from four-node clusters to enterprise-scale environments.
NVIDIA Reference Architectures
Your Guide to the Complete Family
A comprehensive suite of instructions for setting up clusters in the data center is now available.
Use Cases
Designed for Every Use Case
Accelerate agentic AI, physical AI, high-performance computing (HPC), and AI simulation workloads with proven NVIDIA Reference Architectures and NVIDIA-Certified Systems™ from global partners. The primary infrastructure cluster configurations for deploying enterprise AI factories are outlined below.
-
NVIDIA RTX PRO AI Factory
-
NVIDIA HGX AI Factory
-
NVIDIA NVL72 AI Factory
NVIDIA RTX PRO AI Factory
The NVIDIA RTX PRO™ AI Factory configuration is designed for a broad spectrum of enterprise workloads, including generative and agentic AI, data analytics, visual computing, and engineering simulation. Deployments are optimized around 16- and 32-node design points, providing an ideal balance of performance, scalability, and deployment efficiency. Designed for universal workload acceleration across enterprise AI, simulation, and visual computing, NVIDIA RTX PRO Servers are optimized for PCIe environments, making them ideal for space-, power-, and cooling-constrained data centers. Purpose-built for modern AI workloads, they deliver efficient performance for agentic AI and large language model (LLM) inference.
NVIDIA HGX AI Factory
The high-performance NVIDIA HGX™ AI Factory configuration is purpose-built for multi-node AI training and inference at scale, leveraging NVIDIA HGX systems. Available in 32-, 64-, and 128-node design points and supported by NVIDIA Spectrum-X™ networking, the architecture features a flexible, rail-optimized design that enables efficient integration across diverse rack layouts while delivering high-throughput, low-latency performance. It provides breakthrough performance for AI power users running the most demanding workloads, enables large-scale model training and fine-tuning, and dramatically accelerates inference. With next-generation precision and ultra-fast interconnects, the solution achieves up to 15x higher token throughput.
NVIDIA NVL72 AI Factory
The NVIDIA NVL72 AI Factory configuration is designed to train and deploy trillion-parameter models, delivering exascale computing power within a single rack. Built for massive model throughput, multi-user inference, and real-time inference at scale, it enables the next generation of AI-driven innovation. Deployment design points center on four- and eight-rack configurations. Built on a flexible, rail-optimized network, the architecture adapts to diverse rack layouts and system designs while delivering high-bandwidth, low-latency performance. The platform delivers exceptional AI factory output with industry-leading energy efficiency and is powered by fifth-generation NVIDIA NVLink™, FP4 Tensor Cores, and advanced thermal innovations.
Benefits
The Strategic Value of NVIDIA Reference Architectures
Unlock scalable, high-performance AI infrastructure with proven, partner-ready configurations.
Peak Performance for AI Workloads
Meet the intensive demands of AI inference, fine-tuning, and training with architectures that ensure full GPU utilization and performance consistency across multi-node clusters.
Flexible Scaling, Simplified Operations
Easily expand your infrastructure and ensure scalable, streamlined deployment for up to 128 nodes. Build the foundation for full-stack solutions with the NVIDIA Enterprise AI Factory validated design, which leverages our software ecosystem.
Reduce Complexity and TCO
Simplify deployment processes and efficient designs, reduce complexity and total cost of ownership (TCO), while reducing time to value.
Supportability
Follow specific, standardized design patterns to achieve consistent operation from one installation to the next, reduce the need for frequent support, and enable faster resolution times.
Partners
Partnered for Performance
We’re proud to collaborate with leading partners as they bring enterprise AI factory solutions to market. Endorsed designs from these partners have passed our Design Review Board, offering guidance that earns our endorsement in one or more of the following categories: infrastructure, Spectrum-X, and networking logical architecture.
Palantir Sovereign AI OS Reference Architecture With NVIDIA
The Palantir Sovereign AI OS Reference Architecture is based on NVIDIA RAs, tested and qualified to run Palantir’s complete software suite on NVIDIA AI infrastructure with our global system partners. This sovereign AI architecture is critical for customers with latency-sensitive workflows, data sovereignty requirements, and high geographic distribution. The architecture provides enterprises with total control over their data, AI models, and applications.
Resources
Learn More About NVIDIA RAs
NVIDIA IT’s AI Factory Drives Enterprise Innovation at Scale
Leveraging NVIDIA Reference Architectures, NVIDIA built a unified AI factory to scale generative AI and agentic workflows across the enterprise, ensuring security, performance, and consistency. The platform supports hundreds of AI agents that accelerate innovation, streamline software and hardware engineering, and optimize supply chain operations—reducing planning times by over 95% and achieving decades’ worth of engineering work in just one year.
Next Steps
NVIDIA Reference Architecture FAQs
NVIDIA Reference Architectures provide clear, consolidated cluster design guidance for NVIDIA global system partners and enterprise customers building enterprise AI factories. They span compute, networking, storage, and infrastructure software, bringing NVIDIA-Certified Systems, NVIDIA-Certified Storage, NVIDIA Networking, and NVIDIA AI Enterprise software into integrated designs for production AI infrastructure.
Choose RTX PRO AI Factory for broad enterprise AI, including agentic and physical AI, visual computing, analytics, simulation, and large-to-medium model inference.
Choose HGX AI Factory for high-performance multi-node training, fine-tuning, and large-scale inference.
Choose NVL72 AI Factory for the most intensive workloads, including trillion-parameter model training, real-time reasoning, and high-throughput massive-scale inference.
Start with the workload type, number of users, model size, performance target, data volume, and growth plan. NVIDIA publishes design points for each architecture, including 16- and 32-node RTX PRO systems, 32-, 64-, and 128-node HGX systems, and 2-, 4-, and 8-rack NVL72 systems. Work with an NVIDIA systems partner to confirm the design and bill of materials for your deployment.
The code follows a CPU-GPU-network adapter-bandwidth format. In 2-8-5-200, the system has 2 CPU sockets, 8 GPUs, 5 network adapters with 1 for north-south traffic and 4 for east-west traffic, and average 200GbE east-west bandwidth per GPU.
An NVIDIA’s Reference Architecture provides guidance for compute, networking, storage, infrastructure software, and deployment. NVIDIA Enterprise AI Factory Validated Design adds a curated and versioned full stack of NVIDIA and partner software that supports AI workloads. Teams can use the RA to plan AI infrastructure up through the Kubernetes layer and the validated design to deploy and operate the software stack on that infrastructure.
Cisco, Dell Technologies, HPE, Lenovo, and Supermicro offer designs based on NVIDIA RAs and endorsed by the NVIDIA Design Review Board. Check the NVIDIA RA documentation for the current server models, GPU configurations, scale points, and endorsement status before selecting a system.
NVIDIA-Certified Systems validate the node: servers tested to meet NVIDIA requirements for enterprise AI performance, manageability, security, and scalability.
NVIDIA Reference Architectures define the cluster: how NVIDIA-Certified systems, NVIDIA networking, storage, and NVIDIA AI Enterprise software come together in repeatable configurations for Enterprise AI Factories.
In short: NVIDIA-Certified tells customers a system is ready for AI; NVIDIA RAs show partners and customers how to assemble those systems into scalable AI factory deployments.
Cisco is the worldwide technology leader that is revolutionizing the way organizations connect and protect in the AI era. For more than 40 years, Cisco has securely connected the world. With its industry-leading AI-powered solutions and services, Cisco enables its customers, partners, and communities to unlock innovation, enhance productivity, and strengthen digital resilience. With purpose at its core, Cisco remains committed to creating a more connected and inclusive future for all.
NVIDIA Design Review Board-endorsed solutions:
- Cisco Enterprise Reference Architecture for NVIDIA HGX B300: Cisco N9000 | Cisco Nexus Hyperfabric
- Cisco Enterprise Reference Architecture for NVIDIA HGX H200: Cisco N9000 | Cisco Nexus Hyperfabric
- Cisco AI POD Infrastructure for the NVIDIA 2-8-9-400 Enterprise Reference Architecture
- Cisco AI POD Infrastructure for the NVIDIA 2-8-5-200 Enterprise Reference Architecture
Dell Technologies helps organizations and individuals build their digital future and transform how they work, live, and play. The company provides customers with the industry’s broadest and most innovative technology and services portfolio for the AI era.
NVIDIA Design Review Board-endorsed solutions:
- Dell AI Factory With NVIDIA: NVIDIA RTX PRO Servers and H200 NVL
- Dell AI Factory With NVIDIA for Government: NVIDIA RTX PRO Servers
- Dell AI Factory With NVIDIA: HGX H200
HPE is a leader in essential enterprise technology, bringing together the power of AI, cloud, and networking to help organizations achieve more. As pioneers of possibility, our innovation and expertise advance the way people live and work. We empower our customers across industries to optimize operational performance, transform data into foresight, and maximize their impact. Unlock your boldest ambitions with HPE.
NVIDIA Design Review Board-endorsed solutions:
- HPE AI Factory With NVIDIA: NVIDIA RTX PRO Servers
- HPE AI Factory With NVIDIA: NVIDIA HGX B300
Lenovo is a US$69B revenue global technology powerhouse, ranked #196 in the Fortune Global 500, and serving millions of customers every day in 180 markets. Focused on a bold vision to deliver Smarter Technology for All, our ongoing partnership with NVIDIA combines Lenovo servers with accelerated GPUs. The Lenovo Hybrid AI Advantage™ with NVIDIA boosts productivity and innovation with faster AI deployment, powered by the Lenovo AI Library and a full-stack portfolio of AI infrastructure, devices, solutions, and services.
NVIDIA Design Review Board-endorsed solutions:
Supermicro is a global leader in application-optimized total IT solutions. Founded and operating in San Jose, California, Supermicro is committed to delivering first-to-market innovation for enterprise, cloud, AI, and 5G telco/edge IT infrastructure. We are a total IT solutions provider with server, AI, storage, IoT, switch systems, software, and support services. Supermicro’s motherboard, power, and chassis design expertise further enables our development and production, enabling next-generation innovation from cloud to edge for our global customers.
NVIDIA Design Review Board-endorsed solutions:
- AI Factory Solutions With NVIDIA RTX PRO Servers: Datasheet | Reference Architecture
- AI Factory Solutions With NVIDIA HGX B300: Single and Dual Plane: Datasheet | Reference Architecture
- AI Factory Solutions With NVIDIA HGX B200: Datasheet | Reference Architecture