Training on Multiple GPUs

When To Use

Use this skill when a user wants to scale an Isaac Lab RL job across several GPUs or nodes, or when a distributed run hangs, deadlocks, or fails inside NCCL.

Do not use this skill for single-GPU training setup, framework selection, or agent config wiring. Use isaaclab-training-rl-agents for those. Use isaaclab-debugging-rl-training when the job runs to completion but the reward curves or metrics look wrong.

Workflow