Most GPUs are not idle all the time. They're idle some of the time. A gaming PC is busy in the evening and dark the rest of the day. A studio's workstations work office hours. A gaming cafe fills up after school and empties after midnight. A company's GPU cluster is sized for its peaks and quiet between them.
Until now, renting out a machine like that meant a choice: commit it to renters and lose it when you need it, or keep it to yourself and let it sit. Spot nodes remove the choice. The machine earns while it's free and comes back to you the moment you want it.
What a spot node is
A spot node is a GPU node you can take back at any time. It does everything a regular node does on ParalonCloud (it rents to developers, and it runs inference for our API), with three differences:
- One button: Pause or Resume. Available, the network can use it. Paused, nothing new lands on it and whatever is running winds down.
- It rents at a discount. Renters pay the platform default price for its GPUs minus a spot discount, because they're getting a machine that can be taken back. You still keep 80% of what the rental bills.
- Taking it back never costs you reputation. Spot rentals are not counted toward your node's or your provider score.
What happens when you pause
Pausing is designed so that you get the machine back fast and the person using it doesn't lose work they could have saved:
- A rental gets two minutes' notice. The renter sees a countdown on their rentals page and gets a notification, then the rental stops. They pay only until it stops.
- Inference finishes what it's doing. The node stops taking new requests immediately. Requests already running get up to two minutes to complete, then the worker shuts down. Traffic goes to other nodes.
- Nothing new arrives until you resume the node.
Two minutes is the same notice the big clouds give on their spot instances. It's enough for a renter's job to write a checkpoint, and short enough that you can sit down at your machine almost right away.
Who it's for
- Gamers. Rent out your GPU while you're at work or asleep; pause before you play. See Rent Out Your Gaming PC's GPU When You're Not Playing.
- Gaming cafes. Put the floor to work after closing, pause it all with one click before opening. See Gaming Cafe After Hours.
- Design, 3D and VFX studios, and offices. Workstations that sit idle at night and at weekends.
- Data centers and companies with their own GPUs. Sell spare capacity between your own peaks without giving up control of it. See Sell Spare Data-Center GPU Capacity Without Losing Control.
If a machine is dedicated to the network for the long term, keep it a regular node: renters pay more for a machine that won't be taken back, and a clean record builds the reputation that wins rentals.
How to use it
- New machine: in Add node, choose GPU, open Advanced options and tick Spot.
- Machine already connected: on its card, click Edit and tick Spot.
From then on the card shows whether the node is Available or Paused, with a Pause or Resume button. If you run several spot machines, a bar above your node list lets you pause or release all of them at once, or just the ones you've selected.
A node's spot setting can't change while it's rented; it takes effect from the next rental.
For renters
Spot GPUs are marked on the Rent page with the discount and a "Spot · reclaimable" label instead of a reliability score, and there's a filter to show only on-demand or only spot machines. They suit jobs that can stop and resume: training with checkpoints, batch processing, experiments. For a long job that must not be interrupted, pick an on-demand machine. More in Spot vs On-Demand GPUs.
The Rental API reports spot and list_price_per_hour for every GPU, and reclaim_at on a rental once its owner has paused it, so a script can react to the notice.
Start
Everything is in the docs: Spot Nodes for providers, Rent a GPU for renters. If you've been holding back from renting out a GPU because you still need it, this is the version made for you.
