NVIDIA DSX Air: Simulating AI Infrastructure at Scale
NVIDIA has introduced DSX Air, a cloud-based simulation platform designed to help organizations design, test, and validate complete AI factory infrastructure before physical deployment. The service addresses a critical challenge in building large-scale AI systems: the complexity of integrating compute, networking, storage, and security components in a way that ensures optimal performance and prevents costly integration issues.
Key Features
Simulation Capabilities: DSX Air enables end-to-end simulation of AI infrastructure built with NVIDIA Spectrum-X Ethernet, NVLink, Spectrum-6 Ethernet switches, and NVLink switches. This includes the upcoming NVIDIA Vera Rubin platform for next-generation deployments.
Checkpoint & History Management: Users can save snapshots of simulation states to pause and resume work without losing configurations, with checkpoints automatically created when simulations stop. A detailed history log tracks all events throughout a simulation's lifecycle—including state changes, user activities, and errors—with keyword filtering for quick troubleshooting.
Enterprise Collaboration: Integrated with NVIDIA GPU Cloud (NGC), DSX Air provides unified account setup with role-based access controls, allowing organizations to manage multi-user access and resource allocation across teams.
DevOps Integration & Use Cases
The platform includes Python SDK and REST APIs for seamless integration with CI/CD pipelines, enabling automated testing of software updates and configuration changes. This allows teams to continuously verify infrastructure changes without relying on physical hardware. Pre-production validation of provisioning policies, automation workflows, and security configurations helps reduce deployment timelines and integration risks.
Getting Started
DSX Air provides guided demos and training materials covering NVIDIA solutions like Cumulus Linux, Run:ai, Base, and Command Manager. Organizations can replicate production environments to compress deployment cycles and enhance AIOps efficiency.