My laptop vs a rented GPU

The specification proposes a laptop-versus-GPU comparison; this page does not report an executed rental or measured speedup. A fair experiment records hardware, simulator version, network, task, training target, number of transitions and achieved control quality.

Separate compilation and initialization from steady-state sampling and total time to a declared performance target. A fast simulator can produce less useful training data if its dynamics or observation model differ. Isaac Lab and the custom JAX simulator therefore require a model-matching check before their learning curves can be compared meaningfully.

Run the CPU path first and keep it sufficient for the course. If you already have suitable hardware or choose to perform the optional comparison, publish the actual runtime, cost basis and quality reached. Do not infer a guaranteed two-hour run or a rental cost from the planning estimate.