Blueprint Governance

Overview¶

Blueprint governance across organizations enables centralized management of blueprints while allowing controlled reuse across organizations through compute and service profiles.

The default-gpu blueprint is available in all organizations and can be directly used as a base blueprint when creating custom blueprints. However, the sharing scope of custom blueprints depends on where they are created.

Purpose and Applicability¶

This capability provides platform administrators with a controlled mechanism to standardize blueprint usage while supporting centralized or organization-scoped customization.

This applies when:

Base blueprints are used to create custom blueprints
Custom blueprints are created in system-catalog for centralized sharing
Custom blueprints are created in other projects for org-scoped usage
Compute or service profiles reference these blueprints

Default-GPU Blueprint Overview¶

The default-gpu blueprint is a system blueprint designed for Kubernetes clusters using GPU-based worker nodes.

It is available under the Default Blueprints tab along with other system blueprints such as default, minimal, and provider-specific default blueprints.

This blueprint includes all components available in the existing default blueprint, along with additional GPU-specific system add-ons required for GPU workload execution and external access to cluster applications.

The default-gpu blueprint located in the system-catalog project of the default organization acts as the centralized blueprint that can be used during compute and service instance launch across organizations.

Components Included¶

The following add-ons are automatically appended by the system when the default-gpu blueprint is selected. These are managed system add-ons and are not manually configurable through the UI.

Component	Version	Purpose
`gpu-operator`	v24.9.0	Automates GPU enablement in Kubernetes by installing and managing GPU drivers, runtime components, device plugins, and monitoring capabilities required for GPU workloads.
`ingress-nginx`	4.11.3	Provides ingress functionality to expose cluster applications externally using HTTP and related protocols.

GPU Operator Behavior¶

The GPU operator automates the setup required to use NVIDIA GPUs inside Kubernetes workloads.

When installed, the GPU operator:

Installs NVIDIA drivers
Deploys the NVIDIA container runtime
Installs the Kubernetes device plugin
Manages GPU monitoring using DCGM exporter
Supports MIG (Multi-Instance GPU) on compatible GPUs
Discovers GPU resources on worker nodes
Labels nodes based on GPU availability for workload scheduling

Without the GPU operator, Kubernetes cannot identify or expose GPU hardware for workloads.

Ingress Engine¶

The ingress-nginx add-on provides standard ingress functionality for exposing applications running inside the cluster.

Managed System Add-ons Behavior¶

The gpu-operator and ingress-nginx components included with default-gpu are managed system add-ons.

Behavior:

Automatically appended at the backend when default-gpu is used
Cannot be enabled, disabled, or modified through the UI
Not included with other system blueprints

Using Default-GPU in Any Organization¶

The default-gpu blueprint is available in all organizations.

Users can:

Select default-gpu as the base blueprint
Create custom blueprints in any project and any organization
Use those blueprints through compute and service profiles

To create a centralized custom blueprint that can be shared across organizations:

Create a custom blueprint in the system-catalog project under the default-org
Select a base blueprint (for example, default-gpu, default, minimal, or any other blueprint) and extend it with additional add-ons as required
Save changes

Once this custom blueprint is created, it can be used when creating compute or service profiles in any organization.

During blueprint synchronization, the system validates whether the blueprint originates from the system-catalog project under the default-org.

If the blueprint is from the system-catalog project, blueprint synchronization succeeds.
If the blueprint is from any other project or organization, blueprint synchronization fails for cross-organization usage.

This approach allows the custom blueprint to be shared and used across organizations.

Organization-Scoped Custom Blueprint¶

If a custom blueprint is created in a different project under the default organization, or inside any tenant organization:

The blueprint can be shared only to projects within the same organization
It cannot be shared to other organizations

This behavior defines the scope difference between centralized and organization-level blueprint creation.

Blueprint Comparison¶

Blueprint Type	Availability	Sharing Scope
`default-gpu` system blueprint	Available in all orgs	Can be used as base blueprint everywhere
Custom BP in system-catalog (default org)	Centralized	Shareable across all orgs and projects
Custom BP in other projects/orgs	Org-specific	Shareable only within same org

Version Compatibility Consideration¶

The default-gpu blueprint currently includes:

gpu-operator: v24.9.0
ingress-nginx: 4.11.3

GPU operator compatibility depends on worker node operating system versions. If the deployed version is not supported, add-on deployment or blueprint synchronization may fail.

When different versions are required, create a custom blueprint with alternate add-on versions.

Blueprint Configuration Flow¶

Step 1: Create a Custom Blueprint¶

Create a custom blueprint:

In system-catalog (default org) for cross-org sharing, or
In any organization/project for org-scoped usage

Select a base blueprint (for example default-gpu, default, or minimal) and add required custom add-ons.

Step 2: Create a Compute or Service Profile¶

Create a Compute Profile or Service Profile and configure Blueprint Name and Blueprint Version inputs.

Show to User — allows override during profile usage
System Variable — hides blueprint values from users

Share the profile:

Across organizations (only when using system-catalog custom blueprint), or
Within the same organization