Skip to main content
Version: v1.2.x

NVIDIA Run:ai

NVIDIA Run:ai is a GPU orchestration platform that schedules and manages AI and machine learning workloads across Kubernetes clusters. PaletteAI packages Run:ai as a ready-to-use Profile Bundle so platform teams can deploy GPU scheduling on PaletteAI Compute Pools without assembling operators and charts by hand.

Spectro Cloud participates in the NVIDIA Run:ai self-certification program. PaletteAI is listed among the partner-compatible distributions in the NVIDIA Run:ai support matrix. Partner-compatible distributions are tested and validated by the partner, who maintains compatibility with NVIDIA Run:ai.

The Run:ai Profile Bundle is published in PaletteAI Studio. Download it there, import it into your Project through the PaletteAI UI, then deploy it as an App Deployment onto an existing Compute Pool.

Compatibility​

The following versions are validated under the partner self-certification program for Spectro Cloud / PaletteAI:

KubernetesPaletteAINVIDIA Run:ai
1.341.2.02.24

For the latest partner matrix and other distribution details, refer to the NVIDIA Run:ai support matrix — Partner-Compatible Distributions.

info

Exact chart and application versions shipped in a given Profile Bundle revision are defined in PaletteAI Studio (profilebundle.yaml and related workload profiles). Align your Compute Pool Kubernetes version and PaletteAI release with the table above and the bundle you download.

How it fits PaletteAI​

Run:ai is published in PaletteAI Studio as an Application Profile Bundle. Which deployment path you use depends on whether a GPU-backed Compute Pool already exists.

Application path (existing Compute Pool)​

Use this path when a GPU-backed Compute Pool already exists. Deploy the Run:ai Application Profile Bundle through the App Deployment flow onto that pool—the same path used for other application stacks such as ClearML.

PieceRole
Compute PoolProvides the GPU-backed Kubernetes cluster (spoke) where Run:ai runs
Run:ai Profile Bundle (Application)Deploys the Run:ai platform components onto that pool
App DeploymentCreates the workload instance from the Run:ai Profile Bundle

Fullstack path (new Compute Pool)​

If PaletteAI Studio publishes a Run:ai Fullstack variant, it provisions the GPU-backed Compute Pool and deploys Run:ai together. Use the Create and Manage Compute Pools flow to select the Fullstack Profile Bundle—do not use App Deployment to create the pool.

For Profile Bundle types, refer to Types of Profile Bundles.

Get the Run:ai Profile Bundle​

Prerequisites​

  • Access to PaletteAI Studio
  • A Project namespace ready to receive the import
  • For the Application path: a Compute Pool in Running status with NVIDIA GPUs, and a kubectl context for that Compute Pool to run the GPU checks in Validate
  • Only if you import using the CLI: the PaletteAI CLI installed and available in your $PATH, and a kubectl context for the PaletteAI hub cluster (not required for UI-native import)

Download from PaletteAI Studio​

  1. Open PaletteAI Studio.
  2. Search for Run:ai (or NVIDIA Run:ai) in the Profile Bundle catalog.
  3. Open the Profile Bundle and review the description, type (Application / Infrastructure / Fullstack), and version.
  4. Download the Profile Bundle archive (.tar.gz) to your workstation.
tip

If Run:ai is listed as a gated profile, request access through your Spectro Cloud account team, then download after access is granted. For Studio access models, refer to PaletteAI Studio.

Import into PaletteAI​

  1. In PaletteAI, open your Project and select Profile Bundles → Import Profile Bundle.
  2. Upload the Run:ai Profile Bundle archive you downloaded from PaletteAI Studio and select Continue.
  3. Review the preview, then select Finish & Import.
  4. Confirm the imported bundle status is Ready.

For the full UI import workflow, refer to Import Profile Bundles.

If you prefer a command-line workflow, use the PaletteAI CLI:

paletteai studio import --namespace <project-namespace> ~/Downloads/<runai-bundle>.tar.gz

For flag details, refer to paletteai studio import.

Deploy Run:ai​

  1. Log in to PaletteAI and open your Project.
  2. Select App Deployments → Deploy App.
  3. Choose the imported Run:ai Profile Bundle.
  4. Select an existing Compute Pool that meets the bundle's GPU and Kubernetes requirements. If you need a new GPU-backed Compute Pool, provision it first through the Fullstack path instead.
  5. Complete the wizard and deploy.

When the App Deployment is healthy, use any Platform URL or service exposure configured for the bundle to open the Run:ai UI. Refer to Expose a Deployment if you need to publish access.

Validate​

  • Profile Bundle status is Ready in the Project.
  • App Deployment status is healthy / running.
  • GPU nodes on the Compute Pool are schedulable and visible to Run:ai after the stack settles.