Skip to content

Compute

codebyjames edited this page Sep 5, 2026 · 1 revision

Compute

Compute

Whether a job should run on the card in this machine, or on one you rent by the minute.

It answers the same question for both: how long, how much, and does it fit in the card's memory — including the minutes a rented pod spends starting up.

Renting is rarely worth it for a single render and often worth it for a long training run.

Renting

Nothing is rented without a RunPod API key. Without one this screen is an honest comparison that always recommends your own machine.

With a key, ImageForge can provision a pod and tear it down. Everything in between is manual — installing the trainer on the pod, uploading the dataset, launching the run, downloading the adapter.

Billing continues until a pod is TERMINATED, not merely stopped. The app offers a Terminate action and warns about this in several places, because a forgotten idle pod is the single most common way to waste money on rented GPUs.

Clone this wiki locally