NVIDIA DSX OS
Open, modular software for AI factory operations.
Overview
The Operating Layer for AI Factories
Bringing an AI factory into production requires infrastructure and platform teams to connect hardware provisioning, networking, Kubernetes, GPU management, workload scheduling, inference services, security, monitoring, and recovery. Building and operating each layer separately can slow deployment, leave valuable capacity underused, and make failures harder to resolve.
NVIDIA DSX OS™ brings these capabilities together through an open, modular software portfolio. It helps operators bring infrastructure online, maintain a consistent runtime, improve GPU utilization, automate fleet health, and deliver production AI services reliably at scale.
Adopt the components that address immediate requirements, integrate them with your existing environment, and expand as your infrastructure and services grow.
Features
One Modular Software Portfolio for AI Factory Operations
DSX OS provides modular capabilities for bringing accelerated infrastructure online, running AI workloads, and maintaining productive capacity through day-two operations.
Benefits
Improve the Economics of Operating AI Factories
DSX OS helps providers bring infrastructure online faster, increase productive use of accelerated capacity, and operate AI services reliably across multi-tenant environments.
Accelerate Time to Revenue
Reuse modular infrastructure and platform software instead of rebuilding foundational operating capabilities for every environment.
Bring clusters online, validate readiness, and begin delivering AI services faster.
Make More Productive Use of GPU Capacity
Match GPU capacity to workload demand through intelligent scheduling, resource allocation, topology-aware placement, and service orchestration.
Increase Reliability and Resiliency
Maintain runtime consistency, monitor fleet health, identify failures, protect workloads, automate remediation, and return affected capacity to service.
Move operations from reactive alerting toward an automated health-and-recovery lifecycle
Scale Secure Multi-Tenant Operations
Manage infrastructure lifecycles, verify platform integrity, orchestrate resources, and deliver AI services across tenants and clusters.
Adopt individual components or combine capabilities as operational requirements grow.
Starting Options
Start With the Capability Your AI Factory Needs
DSX OS is designed for incremental adoption. Begin with a focused operational requirement and expand over time.
Resources
The Latest From the NVIDIA DSX Ecosystem
Next Steps
FAQs
NVIDIA DSX OS™ is an open, modular software portfolio for building, operating, and scaling multi-tenant AI factory infrastructure. It includes open-source components and related NVIDIA technologies for platform and infrastructure operations.
Bringing an AI factory into production requires infrastructure and platform teams to coordinate hardware provisioning, networking, Kubernetes, GPU management, workload scheduling, inference services, security, monitoring, and recovery. Operating these layers separately can slow deployment, leave capacity underused, and complicate failure recovery. DSX OS provides modular technologies that can be integrated into existing platforms and software stacks.
DSX OS is modular and designed for incremental adoption. Organizations can start with the components that address their immediate operational requirements, integrate them with existing software stacks, and add capabilities as their infrastructure and services grow.
DSX OS platform software includes KAI Scheduler, Karta, NVIDIA Run:ai, NVIDIA Cloud Functions (NVCF), Grove, and NVIDIA Dynamo. These components support workload scheduling and GPU orchestration, service orchestration, and distributed inference.
DSX OS infrastructure software includes NVIDIA GPU Operator, NVIDIA Network Operator, Topograph, NVIDIA AI Cluster Runtime, NVIDIA Infra Controller (NICo), NVIDIA DOCA Platform Framework (DPF), NVIDIA Switch Infrastructure - Config Manager, and NVIDIA Rack Management Service. Together, these components support cluster runtime, topology awareness, bare-metal provisioning, multi-tenant lifecycle operations, and rack and network infrastructure management.
NVIDIA Cluster Readiness Engine supports cluster burn-in certification, orchestrated benchmarking, and hardware-failure detection. NVIDIA Fleet Intelligence provides real-time insights into GPU fleet health and integrity. NVIDIA NVSentinel provides Kubernetes-native GPU monitoring and fault remediation.
Implementation guidance, quick starts, reference designs, and links to open-source projects are available in the NVIDIA DSX documentation.