AICR selected and ordered a GPU training platform. ConfigHub stores the 17 exact Argo CD Applications produced from that recipe as one base variant.
Start here when you open this Space in Hub. This page explains the problem this example is meant to show, what to inspect, why it matters, and where the evidence lives.
Why this example exists
AICR can choose and package the parts of an AI platform, but a platform team still needs a record of which recipe, package, and generated configuration each cluster should run.
This example keeps the AICR recipe and OCI digest connected to the exact Argo CD objects in ConfigHub. Teams can review the objects, create environment or cluster-class variants, and promote changes without rebuilding the package from memory.
What this example shows
- The AICR v0.14.0 recipe selected 15 versioned components for EKS, H100 accelerators, Ubuntu, Kubeflow, and training.
- The generated Argo CD configuration contains one parent Application and 16 component Applications, ordered with sync waves 0 through 15.
- ConfigHub imported those 17 Applications from one OCI configuration artifact without running AICR or rendering the source chart again.
- This Space requires approval before apply because it changes cluster-wide GPU, monitoring, and training-platform configuration.
Where each setting comes from
| Place | What this Space records |
|---|---|
| Starting configuration | Open the upstream Space or source link named on this page. That is the configuration this Space started from. |
| ConfigHub changes | Open Unit revision history. It records changes made after the starting configuration was saved. |
| Install work | Use the linked chart or route evidence for required Secrets, CRDs, hooks, setup jobs, and target facts. |
| Live cluster | Use target observations to compare the cluster with the reviewed Units. A live-only edit is drift until it is recorded or removed. |
If an upstream change and a local ConfigHub revision touch the same field, review the overlap before promotion.
What to inspect in Hub
- This README.
- The
aicr-eks-h100-training-kubeflowUnit to inspect the 17 Applications and their sync waves. - The Space annotations to see the OCI source reference and resolved digest.
Evidence and source
- AICR example guide
- AICR source and OCI receipt
- Public OCI receipt
- ConfigHub upload receipt
- Development and staging promotion
- Apply policy and live assignments
- Rendered Argo CD Applications
Limits
- This proves the package-to-base-variant path. It does not claim that Argo CD reconciled the Applications or that the workloads ran on an EKS GPU cluster.
- The source Helm chart OCI and the 17-object literal configuration OCI are public and anonymously pullable. The ConfigHub Space records the public literal OCI and its digest.
- The development and staging Spaces prove one reviewed Application change and promotion. They do not prove an AICR package upgrade.
- The target must already provide the
argocdNamespace, the default Argo CD AppProject, Argo CD itself, EKS, and the required GPU capacity.