Add a RunPod GPU pod to your SSH configuration, then select it as a compute target in the local orx dashboard.
In the RunPod dashboard, open Settings > SSH public keys and add your computer's public key. RunPod applies it to new pods.
cat ~/.ssh/id_ed25519.pub

Open the pod's Connect tab and copy the SSH over exposed TCP hostname and port.

Important: Use the exposed TCP connection, not the ssh.runpod.io proxy.
Add the pod to ~/.ssh/config using the hostname and port from RunPod:
Host runpod-gpu HostName <hostname> User root Port <port> IdentityFile ~/.ssh/id_ed25519
ssh runpod-gpu
Run orx up, then open Settings > Compute > SSH. Test runpod-gpu and make it the default compute target if desired. The machine needs git and bash.