Skip to navigation

NVIDIA DSX™

View as Markdown
NVIDIA DSX™

NVIDIA DSX platform brings together reference designs, APIs, software libraries, and technologies to design, simulate, build, and operate AI factories. Explore the features and capabilities for each DSX component, and view linked technical resources.

Select a feature/capability tile to view linked resources, select a section title to jump to its detailed tab below.

DSX Sim™
DSX OS™
DSX MaxLPS™
DSX Exchange™
DSX Hardware
DSX Facilities Infrastructure
DSX Flex
Grid Operator

DSX Sim

A suite of simulation technologies that together enable partners to design, validate, and operate AI factories before and after physical deployment.

NVIDIA DSX Air
High-fidelity logical simulation of NVIDIA hardware and software infrastructure — GPUs, SuperNICs, DPUs, switches — plus validated integrations with storage, security, and orchestration partners.
NVIDIA Omniverse DSX Blueprint for AI Factories
An end-to-end framework to design, simulate, build, and operate gigawatt-scale AI factory digital twins.

Browse the DSX Blueprint GitHub repo or get started on build.nvidia.com
AI Factory Digital Twin Pipeline Samples
Sample scripts and presets for creating DSX SimReady USD assets, covering CAD ingestion, optimization, validation, and metadata workflows for digital twin and AI factory applications.

Browse the AIF Pipeline Samples GitHub repo

DSX OS

Open-source, modular software for building, operating, and scaling multi-tenant AI factory infrastructure with efficiency, performance, and reliability.

Model Serving

Deploys models at scale as production AI inference services optimized for performance and efficiency.

Grove
Modular component of NVIDIA Dynamo — provides a Kubernetes API for defining and scaling multi-component AI inference workloads.
Dynamo
Distributed inference-serving framework built to deploy models in multi-node environments.

Browse the Dynamo GitHub repo

Workload and GPU Orchestration

Allocates and shares GPU capacity efficiently across complex, distributed AI workloads.

NVIDIA Run:ai
Kubernetes-native platform for AI workload and GPU orchestration, designed to maximize accelerated infrastructure utilization.

Browse the Run:ai GitHub repo
NVIDIA Cloud Functions (NVCF)
Unified API layer for scaling inference and simulation workloads across one or more Kubernetes clusters.

Browse the NVCF GitHub repo
KAI Scheduler
Open source Kubernetes-native scheduler for AI workloads with topology-aware placement and resource allocation.
Workload Map
Translation layer that maps any Kubernetes framework's Custom Resource Definitions (CRDs) into a standardized, generic structure.

Health and Resilience

Defines, identifies, contains, and helps recover from workload and infrastructure issues.

NVIDIA Health Validation
Hardware health check catalog and evaluation engine for NVIDIA GPU nodes.

Browse the Health Validation GitHub repo
NVIDIA Attribution
Service and blueprint that correlates logs, events, and metrics from failed GPU workloads into structured diagnoses with symptom chains and root causes.

Browse the Attribution GitHub repo
NVIDIA Recovery
Detects crashes, hangs, and stragglers in distributed AI training and coordinates automated recovery.

Browse the Recovery GitHub repo
NVSentinel
Open-source, Kubernetes-native GPU monitoring and fault remediation.

Browse the NVSentinel GitHub repo
NVIDIA Cluster Readiness Engine
A Kubernetes controller for GPU cluster burn-in certification, orchestrated benchmarking, and hardware failure detection.

Browse the NVCRE GitHub repo

Security and Confidential Computing

Establishes trust and protects AI workloads, data, and infrastructure.

NVIDIA Attestation
The NVIDIA Attestation Suite enhances Confidential Computing by providing robust mechanisms to ensure the integrity and security of devices and platforms.

Observability

Provides visibility into AI factory health, performance, and operations.

NVIDIA Fleet Intelligence
Agent-based managed service offering real-time insights into GPU fleet health and integrity.

Browse the Fleet Intelligence Agent GitHub repo

Accelerated Kubernetes Foundation

Defines, configures, validates, and deploys accelerated Kubernetes environments.

AI Cluster Runtime
Canonical, continuously validated definition of the NVIDIA-accelerated Kubernetes runtime.

Browse the AICR GitHub repo
NVIDIA GPU Operator
Automates the management of NVIDIA software components required to provision GPUs in Kubernetes.

Browse the GPU Operator GitHub repo
NVIDIA Network Operator
Automates the provisioning and management of NVIDIA networking resources in Kubernetes clusters.

Browse the Network Operator GitHub repo
Nodewright
A Kubernetes-aware package manager for cluster administrators to safely modify and maintain underlying hosts declaratively at scale.

Browse the Nodewright GitHub repo
NVIDIA Kubernetes Engine (NVKE)
Management plane for creating and operating managed Kubernetes clusters on NVIDIA GPU infrastructure.

Browse the NVKE GitHub repo
NVIDIA Kubernetes Recipes (NVKR)
Build system and declarative recipes for producing NVKE node images, packages, and components from pinned inputs.

Browse the NVKR GitHub repo
Topograph
Discovers the physical network topology of a cluster and exposes it to schedulers, enabling topology-aware scheduling decisions.

Browse the Topograph GitHub repo

Infrastructure Operations

Provisions, manages, and operates AI factory infrastructure and compute.

NVIDIA Infra Controller (NICo)
Bare-metal provisioning and secure lifecycle management for multi-tenant GPU infrastructure.

Browse the NICo GitHub repo
NVIDIA DOCA Platform Framework (DPF)
Orchestration system to build, deploy, and operate BlueField-accelerated infrastructure services.
NVIDIA Switch Infrastructure - Config Manager
Open-source network automation and configuration management platform for large-scale data center operations.

Browse the Config Manager GitHub repo
NVIDIA Rack Management Service
A stateless, rack-level hardware management service that trusted services and operators use to run power control, inventory, firmware updates, and switch configuration for NVIDIA infrastructure.

Browse the RMS GitHub repo

DSX MaxLPS

NVIDIA DSX MaxLPS is a suite of chip, thermal, system, and software technologies that maximizes AI factory performance-per-watt within a fixed power envelope.

DSX MaxLPS Overview
Explore the core MaxLPS concepts, see how they work together within a fixed power envelope, and find next steps for applying them to AI factory designs.
Sizing DSX AI Factory Infrastructure for MaxLPS
Reference Design Technical Note on infrastructure-sizing considerations across MaxP and MaxQ GPU power profiles and the MaxLPS AI factory framework. (NVOnline #1161311)
NVIDIA Dynamic Power Software
For approved early-preview engagements, learn the dynamic power-allocation model and use the documentation to scope a site pilot for GPU and rack power management.

DSX Exchange

The IT/OT integration and communications hub that enables compute, network, energy, power, and cooling plant systems to share operational signals across the AI factory.

DSX Event Bus
NATS-based event bus for AI factory communications and operations.

Browse the DSX Exchange GitHub repo
AsyncAPI Schemas
Event bus schemas for DSX data exchange services.

Browse the AsyncAPI Schemas GitHub source
Agent Gateway
The external entry point that operator agents (run by DSX operators) and tenant agents (run by tenants consuming DSX compute) use to reach DSX MCP servers.

Browse the Agent Gateway GitHub source
BMS Companion Guide
This guide is for System Integrators and BMS contractors who will configure a Building Management System to publish facility data to the DSX Event Bus. It covers the concepts, topic structure, metadata fields, and implementation steps needed to complete a BMS integration.

DSX Flex

DSX Flex enables renewable and hybrid power orchestration across utility, on-site renewables, and storage, providing a schema to receive grid signals (load shedding, demand response, pricing events) and dynamically adapt AI workloads.

NVIDIA DSX Flex Overview
Learn how DSX Flex connects grid-facing requirements with AI factory power management, who participates, and how it works with DSX Exchange and MaxLPS.
NVIDIA DSX Flex Architecture
Understand the participant roles, event lifecycles, and system boundaries defined by NVIDIA DSX Flex.
NVIDIA DSX Flex Grid Services Integration Guide
Connect a grid services solution to DSX Flex, from a grid-event power-limit request through the reported response of managed loads.
DSX Flex Power Management Schema
AsyncAPI schema for integrating with DSX Exchange, providing messages for grid load targets, power states, power-breach alerts, and enforcement outcomes.

DSX Reference Designs

NVIDIA DSX Reference Designs provide generation-specific, validated AI factory architectures.

Explore related facilities infrastructure guidance, power-system qualification resources, and validated ecosystem products for AI factory deployment.

(NVOnline access required. Contact your NVIDIA representative for details.)
NVIDIA Vera Rubin NVL72 Reference Design
(NVOnline #1151654)
NVIDIA DSX - Vera Rubin Facilities Infrastructure Reference Design
(NVOnline #1145739)
NVIDIA DSX Facilities Infrastructure Design Guide
(NVOnline #1152370)
DSX Facilities Infrastructure Reference Design Overview
Interactive walkthrough of the components and technologies within the DSX Facilities Infrastructure Reference Designs.
BESS Self-Qualification Guidelines
Defines the partner-run qualification process for a Battery Energy Storage System (BESS) intended to support AI load buffering, demand response (DR), and low voltage ride-through (LVRT) in grid-connected and islanded on-site generation modes.
NVIDIA DSX AI Factory Marketplace
Products that have been validated to meet NVIDIA DSX Ready functional requirements for AI factory applications.

NVIDIA Cloud Partner Resources

NVIDIA Cloud Partner Software Reference Guide
Infrastructure-native reference for building AI cloud services — software components, architecture patterns, and deployment guidance.
NVIDIA Inference Reference Architecture
Software architecture to help operators build performant, cost-effective inference solutions on NVIDIA platforms.
NVIDIA Requirements for AI Clouds
Primary requirements document for NVIDIA to cloud partners (NCPs) providing compute and software services, defining service delivery expectations, and providing a reference set of requirements and capabilities.
Inference Provider Platform Requirements on GB300 NVL72 for NCPs
Requirements for an NVIDIA Cloud Partner (NCP) building a GB300 NVL72 compute cluster to deliver a managed Kubernetes cluster to Inference Service Providers (ISPs), minimizing friction for deploying and running inference platforms.
NVIDIA DSX Ready
A trusted program for AI factory builders, offering products and solutions that meet category-specific NVIDIA DSX requirements.
NVIDIA Exemplar Cloud
Improves performance per TCO, security, and reliability for cloud providers with hardware and software recipes, tools, capabilities, and references.

Browse the Exemplar Performance recipes
NVIDIA AI Cloud Ready
Qualifies and validates AI infrastructure and platform software for deployment on NVIDIA-accelerated cloud infrastructure.

Browse the NVIDIA AI Cloud Validation Suite