Compute Resources
Compute resources are how PaletteAI turns your physical machines into Kubernetes clusters that run AI/ML applications and models.
-
Compute — An automatically maintained inventory of the machines PaletteAI discovers through Palette, including each machine's CPU architecture, memory, and GPUs.
-
Compute Pools — Kubernetes clusters provisioned from machines in the Compute inventory. Applications and models are deployed to Compute Pools.
-
Compute Configs — Reusable blueprints of cluster settings, such as control plane size, networking, and SSH keys. Apply a Compute Config when creating a Compute Pool instead of re-entering the same values each time.
-
Scaling Policies — Autoscaling rules that you attach to a Compute Pool to automatically add or remove machines based on CPU and GPU utilization.
-
Network Isolation — Per-tenant separation of a Compute Pool's GPU and ingress traffic on shared AI edge infrastructure, using an external network fabric provider. Storage can be isolated too or remain shared, and the management path stays shared by design.
How the Pieces Fit Together
The Compute Pool is the central resource: it draws machines from the Compute inventory, takes its defaults from a Compute Config, scales according to a Scaling Policy, and hosts your App and Model Deployments.
A typical workflow looks like this:
- Machines are registered in Palette and appear in the Compute inventory, which you can review at the tenant scope under Tenant Settings > Compute.
- An administrator creates a Compute Config with the cluster defaults for the Tenant or Project.
- A user creates a Compute Pool, optionally applying the Compute Config and a Scaling Policy. PaletteAI selects matching machines from the inventory and provisions the cluster through Palette.
- Data scientists deploy applications and models to the running Compute Pool.
Next Steps
- Compute Pools — Understand pool types, provisioning, and autoscaling.
- Create and Manage Compute Pools — Create your first Compute Pool.
- Create and Manage Scaling Policies — Automate scaling for a Compute Pool.