Chapter 7 · OperationsGTC Taipei; expanded at GTC Berlin, 21 October 2026
NVIDIA DSX OS: open-source software for running AI factories, and a cluster readiness engine
NVIDIA
A set of open-source, modular components for bare-metal management, configuration control, fault handling and multi-tenancy in GPU clusters. In October NVIDIA added a Cluster Readiness Engine that tests a cluster before it goes into production.
What was announced?
NVIDIA describes DSX OS as “open source, modular software components” for operating and scaling multi-tenant AI factories. It is not one operating system in the usual sense: it is a collection of components that an operator can adopt one at a time.
Components named by NVIDIA include NVIDIA Infra Controller (bare-metal management through an API, with tenant isolation using BlueField DPUs, the data processing units on each server), AI Cluster Runtime (fixed, tested versions of the software stack), NVSentinel (GPU fault detection and automatic draining of failing nodes), and the KAI Scheduler, Dynamo and Fleet Intelligence. On 21 October NVIDIA added three components in developer preview: Config Manager (network configuration), Rack Management Service (rack power, inventory, firmware and switch management) and the Cluster Readiness Engine.
Why does it matter?
A GPU cluster is only useful when every node, link and software version works together. Commissioning one today means stitching together vendor tools, scripts and spreadsheets, and failures often surface only when the first large job runs.
The Cluster Readiness Engine runs distributed test workloads across the cluster before it is handed over and isolates failures to specific nodes. That turns “is the cluster ready?” from a judgement into a repeatable test, and NVIDIA says NVSentinel drains failing workloads “in seconds rather than minutes or hours”.
What has changed?
| Aspect | Typical today | DSX OS approach |
|---|---|---|
| Cluster acceptance | Burn-in scripts and benchmark runs, often custom per site | A readiness engine that runs distributed workloads and names failing nodes |
| Configuration drift | Versions diverge node by node over time | A pinned, tested runtime configuration |
| GPU faults | Alert, then a human drains the node | Automatic detection and drain (NVSentinel) |
| Licensing | Mixed: vendor tools and in-house scripts | Open source, per NVIDIA |
How does it work?
Think of DSX OS as the layer between the hardware and the schedulers. Infra Controller brings a bare-metal server under API control and assigns it to a tenant; AI Cluster Runtime installs a known-good combination of drivers, libraries and operators; Rack Management Service handles power, inventory and firmware at rack level; Config Manager pushes network configuration.
Once the cluster is built, the Cluster Readiness Engine runs workloads that exercise GPUs, memory and the network together, the way a real training job would, and reports which nodes or links fail. In production, NVSentinel watches for GPU faults and moves work off a failing node automatically. NVIDIA’s DSX MaxLPS and DSX Flex power software plug into the same layer.
Figures and claims, and what backs them
- Manufacturer claimNVIDIA says NVSentinel drains workloads from a failing node “in seconds rather than minutes or hours”.
- Published specificationNVIDIA states the components are open source and available on GitHub; Config Manager, Rack Management Service and the Cluster Readiness Engine are in developer preview.
Impact on AI infrastructure delivery
- IT Infrastructure
- Commissioning gets a defined exit test. The IT infrastructure TPM can write acceptance criteria against the readiness engine’s results instead of a negotiated benchmark, and plan time to fix the nodes it isolates.
- Networking
- Network configuration becomes a software artefact (Config Manager) that must be reviewed and version-controlled like code.
- Operations
- Automated drain and pinned configurations reduce manual work, but the operations team owns a new open-source stack: upgrades, security patches and support arrangements have to be planned, since developer-preview components carry no production guarantee.
Commercial availability
- Announced (current stage)Described publicly; no date for customers to buy it.
- PlannedA stated availability window that has not arrived yet.
- ShippingIn production and reaching early customers or partners.
- Generally availableOrderable by any customer through normal channels.
- DeployedRunning in customers’ production environments, confirmed by them or the manufacturer.
NVIDIA says the software is being released as open source and is available on GitHub; the three October components are “in developer preview”. No source states general availability for DSX OS as a whole. CoreWeave, Crusoe, Lambda, Nebius, Nscale and others are described by NVIDIA as “exploring” the new components.
Not yet confirmed
- Which components are actually published on GitHub, and under which licence, was not checked.
- No independent results for the readiness engine or NVSentinel.
Learn more in BSS/OSS Academy
Official sources
- European Ecosystem Builds World’s Most Efficient AI Factories With NVIDIA DSX(opens in a new tab)NVIDIA Newsroom · 21 October 2026 · Manufacturer
- NVIDIA DSX OS Delivers Open, Modular Software for Operating AI Factories at Scale(opens in a new tab)NVIDIA Technical Blog · 31 May 2026 · Technical documentation
- NVIDIA DSX Gives Infrastructure Builders the Playbook for AI Factories(opens in a new tab)NVIDIA Newsroom · 31 May 2026 · Manufacturer
Figures are as the manufacturer states them unless marked otherwise. None has been independently tested by BSS/OSS Academy.