Keep 100% of every rental.
Install one agent on a Linux machine with NVIDIA GPUs. You set the price, we verify the hardware, handle renters and billing, and pay you every Friday. You pay a flat monthly plan instead of a cut.
curl -fsSL https://rent-gpu.app/install.sh | sudo bash -s -- --token rgi_…
Ubuntu 22.04 or 24.04, or Debian 12, with the NVIDIA driver installed. No port forwarding or static IP needed: the agent connects out to us.
Estimate your monthly earnings
- Rental income
- $164.25
- Hobby plan
- −$9.00
- You keep
- $155.25/month
No live RTX 4090 listings yet; the starting price is an estimate. The first 30 days of any plan are free.
Plans
| Plan | Listed GPUs | Per month | Relay traffic included | Also includes |
|---|---|---|---|---|
| Hobby | 2 | $9.00 | 500 GB | console, verification, weekly payouts |
| Pro | 10 | $29.00 | 2 TB | host api, webhooks, bulk editor, price coach |
| Fleet | 50 | $99.00 | 5 TB | host api, webhooks, bulk editor, price coach, priority support, maintenance api |
| Datacenter | unlimited | $299.00 | 25 TB | host api, webhooks, bulk editor, price coach, priority support, maintenance api, datacenter tier eligible, dedicated support |
Plans are charged from your earnings first. If earnings don't cover it, we use your account credit. Unpaid plans pause listings after 7 days; running rentals continue.
What verification checks
The checklist runs when your machine connects. Then one hourly proof runs on idle GPUs, never during a rental. Your machine page shows each measured value next to its threshold.
| GPU identity | every GPU's UUID and PCI ID match a known model, with no virtual or spoofed devices |
| VRAM | 90% of each idle GPU's memory is written and read back |
| Compute | an FP16 matrix multiply reaches at least half the model's rated throughput |
| PCIe | host-to-GPU copies reach at least 60% of the link's rated bandwidth |
| Disk | at least 100 GB free and 150 MB/s sustained |
| Network | at least 100 Mbps down and 50 Mbps up, measured by our relay, under 150 ms away |
| Isolation | Docker with seccomp, cgroup v2 and the NVIDIA Container Toolkit |
| Payouts | identity verified with our payout provider |
Machines then complete a probation period of online time with passing proofs. Reliability is published with its inputs: uptime, failed starts, unscheduled downtime and failed proofs over 30 days. Scheduled maintenance, announced 24 hours ahead, doesn't count against you.