Red Hat maps OpenShift integration points across NVIDIA DSX OS
The joint architecture connects bare-metal provisioning, validated cluster recipes, GPU remediation and MCP-based operations to OpenShift controls.
Red Hat has described how it is integrating OpenShift and Red Hat AI with components of NVIDIA’s DSX OS, laying out an operating model for multi-tenant GPU infrastructure that spans provisioning, configuration, health remediation and agent-driven operations.
The Red Hat engineering post, published Aug. 7, is more architectural than a product-release announcement. Its significance is the specificity of the integration points: the companies are not presenting DSX as a separate appliance beside OpenShift, but as infrastructure services exposed through OpenShift’s control plane and security boundaries.
Provisioning and configuration become cluster APIs
Red Hat says NVIDIA Infra Controller, or NICo, integrates with OpenShift to make bare-metal lifecycle management API-driven. NICo uses NVIDIA’s DOCA Platform Framework and BlueField DPUs for provisioning and hardware-enforced tenant isolation; Red Hat says those boundaries can be managed from the OpenShift control plane.
The architecture also connects NVIDIA AI Cluster Runtime to OpenShift. AICR packages validated runtime configurations as versioned recipes, while Red Hat supplies Helm-template-style integration for components including Node Feature Discovery and NVIDIA’s GPU and Network Operators. The intended result is repeatable configuration across deployments rather than locally assembled GPU stacks.
NVIDIA’s broader DSX OS description confirms that NICo handles programmable bare-metal lifecycle operations and that AICR captures version-locked runtime configurations. NVIDIA lists Red Hat among the companies integrating those components.
Automated GPU remediation
For fleet health, Red Hat says it has validated NVIDIA NVSentinel on OpenShift. NVSentinel detects GPU hardware or driver faults, then cordons affected nodes and moves workloads. NVIDIA Fleet Intelligence agents provide aggregated health and utilization telemetry, which Red Hat says can be surfaced in Grafana dashboards across hybrid environments.
The post also describes a pre-deployment path: Single Node OpenShift can run inside NVIDIA DSX Air so teams can simulate network topologies, BGP routing and Day 2 upgrade procedures before changing physical infrastructure.
MCP enters the operations layer
The most forward-looking integration concerns Model Context Protocol. NVIDIA says DSX OS components will expose MCP servers for provisioning, networking and observability. Red Hat’s architecture places those servers behind OpenShift namespace boundaries, RBAC and audit logging so agents can query telemetry and correlate faults without receiving unrestricted administrative access.
That is a design claim rather than evidence of a generally available autonomous-operations product. The practical development is narrower and useful: Red Hat and NVIDIA have now documented where DSX lifecycle, health and agent interfaces meet OpenShift’s existing control and governance model.
sources
comments · 0