Skip to content

Add baremetal deployment support for NVIDIA GPUs and RDMA/RoCE networking#159

Open
bbenshab wants to merge 2 commits into
redhat-ai-services:mainfrom
bbenshab:feature/add-baremetal-support
Open

Add baremetal deployment support for NVIDIA GPUs and RDMA/RoCE networking#159
bbenshab wants to merge 2 commits into
redhat-ai-services:mainfrom
bbenshab:feature/add-baremetal-support

Conversation

@bbenshab

Copy link
Copy Markdown

This PR adds baremetal deployment support for Red Hat OpenShift AI with NVIDIA GPUs and RDMA/RoCE networking.

Changes

  • Baremetal overlays for rhoai-stable-2.25, rhoai-fast, and rhoai-eus-2.16
  • SR-IOV Network Operator deployment and configuration
  • NVIDIA Network Operator with MOFED drivers for RDMA
  • NVIDIA GPU Operator with RDMA support
  • Automatic RoCE NIC discovery and SR-IOV policy generation
  • Node preparation with automatic labeling
  • GitOps health monitor with auto-cleanup capabilities
  • SR-IOV webhook readiness checks and retry logic
  • Deployment documentation

Testing

  • Successfully deployed on baremetal cluster with ConnectX-5 Ex NICs
  • All operators deployed and healthy
  • SR-IOV policies created automatically
  • No manual intervention required after initial deployment

@bbenshab bbenshab requested a review from a team as a code owner January 22, 2026 12:35
@bbenshab bbenshab force-pushed the feature/add-baremetal-support branch 7 times, most recently from 79e6b25 to 1f68b9c Compare January 22, 2026 12:54
…king

- Baremetal overlays for rhoai-stable-2.25, rhoai-fast, and rhoai-eus-2.16
- SR-IOV Network Operator deployment and configuration
- NVIDIA Network Operator with MOFED drivers for RDMA
- NVIDIA GPU Operator with RDMA support
- Automatic RoCE NIC discovery and SR-IOV policy generation
- Node preparation with automatic labeling
- GitOps health monitor with auto-cleanup capabilities
- SR-IOV webhook readiness checks and retry logic
- Deployment documentation
@bbenshab bbenshab force-pushed the feature/add-baremetal-support branch from 1f68b9c to 49fb9f4 Compare January 22, 2026 13:09
NFD (Node Feature Discovery) is required for GPU and network operators
to detect hardware features on nodes.

Changes:
- Add NFD operator and instance baremetal overlays
- Add NFD to all baremetal ApplicationSets (eus-2.16, fast, stable-2.25)
- Deploy NFD at wave 0-1 (before network and GPU operators)

NFD labels nodes with hardware features including:
- Mellanox NICs (feature.node.kubernetes.io/pci-15b3.present)
- NVIDIA GPUs (feature.node.kubernetes.io/pci-10de.present)

Without NFD, NVIDIA Network Operator and GPU Operator cannot detect
nodes with compatible hardware.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant