Resources step

The first wizard step is Resources. Choose where training runs: how many nodes and which compute class Arena lists for your organization.

What you see

The page heading is Select a training cluster. Copy below it explains that the cluster affects later training settings and that you can change it again while setting up the experiment.

Two controls:

• Number of nodes: buttons for 1, 2, 4, 8, 16, and 32 (shown as 1 nodes, 2 nodes, and so on). Helper text notes that more nodes cost more but can improve performance.

• Resource class: radio list of classes (each option shows node count, hardware details, and price). The Edit resources modal uses a Resource class dropdown instead.

On Advanced Training projects, only GPU resource classes are listed on Resources from the start. Classic reinforcement learning projects can use CPU-only classes as well.

Arena pre-selects the first enabled class when you open a new draft.

Moving forward

Next stays clickable, but if you have not chosen a resource class you get a toast: Please choose a preferred compute resource.

Save appears only on Summary (with Train or Resume). Earlier steps persist when you click Next or after idle autosave. A resource class is still required before Train on Summary.

On Summary, Train only appears when the experiment is ready to schedule (including a chosen resource). The tooltip on Train says training usually starts within about ten minutes once resources are ready.

Changing resources later

On steps after Resources, use Edit resources to open Select training cluster. That modal has Number of nodes, Resource class, and Save.

After this step

Environment is next. Classic projects go straight to gym selection. Advanced Training projects pick a gym-style RL Environment or a Dataset from the combined grid.

See Environment step.