Host: earn with your PC
Step by step from download to first payout: install the host app, sign in with a code, set your price, pick an interruption policy, understand idle-only, electricity and payouts.
Turn an idle gaming PC or workstation into a GPU other people pay to use. You set the price and the rules; the platform keeps a flat fee and you keep the rest. About ten minutes, most of it downloads.
Before you start: renters' code runs in Docker on your PC. Use a machine you do not keep secrets on, never a work laptop.
Requirements
| Minimum | Notes | |
|---|---|---|
| GPU | NVIDIA GTX 1660 / RTX 2060 or newer, 6 GB VRAM | 24 GB cards (3090 / 4090 / 5090) earn the most. AMD / Intel are not supported yet. |
| OS | Windows 10 21H2+ or Windows 11, 64-bit; or Ubuntu 22.04+ | macOS cannot host (no NVIDIA). |
| RAM / disk | 16 GB RAM, 60 GB free on an SSD | Base images are 8–15 GB; renters' images more. |
| Internet | 20 Mbps upload | Upload matters more than download. Unlimited data plan recommended. |
| Ports | none | The app only connects outbound. No router changes. |
Step 1: install and sign in
Windows
- Go to Host and click Download FairGPU Host for Windows (sign up first if you have no account).
- Run the installer. The app sits in the system tray and starts with Windows.
- Sign in with the code. The app shows a code like
ABCD-1234and opens the site at Approve sign-in. Check the device name and click Approve; you never type a password into the app. - Docker is installed for you. The app checks for WSL 2 and Docker Desktop and installs what is missing. If WSL 2 had to be enabled, Windows asks for one reboot; the app continues afterwards. You only need an up-to-date NVIDIA driver.
- Within a minute the PC appears in your dashboard as ONLINE with its GPU detected.
Linux
There is no desktop app for Linux; the agent runs as a systemd user service.
- Install the NVIDIA driver, Docker Engine and the NVIDIA Container Toolkit; verify with
docker run --rm --gpus all nvidia/cuda:12.4.1-base-ubuntu22.04 nvidia-smi. - Dashboard → Add machine → Advanced: install without the app, name the machine, copy the one-line installer (it carries a one-time token):
curl -fsSL https://fairgpu.io/api/agent/install.sh | bash -s -- --token fg_xxx --server https://fairgpu.iosystemctl --user status fairgpu-agentshows it running; the dashboard shows the machine ONLINE within a minute.
macOS
Hosting needs an NVIDIA GPU, so macOS cannot host. You can still rent from a Mac: see Rent your first GPU.
The dashboard's Setup checklist ticks each step live: app installed, signed in, Docker ready, price, policy, availability, first rental, payout method.
Step 2: set your price
Machine → Pricing. You see the market median and range for your GPU and a suggestion; type any number. The price is for the whole machine. Renters see this price; the platform fee comes out of it, so you receive 90 %.
Also here: smart pricing (nightly ±5 % within bounds you set), daily / weekly discounts. Under Settings: minimum rental (default 10 min), max rental hours (set it to what fits your evenings), allow interactive / jobs.
Step 3: choose your interruption policy
Machine → Booking → If you need the PC back. This is shown on your listing and renters filter on it.
- Committed: you never take the PC back during a rental. If the rental still ends on your side (PC off, crash, Docker down), the renter is refunded in full and you earn nothing for it.
- With notice (10–120 min): Reclaim now asks for at least that notice; the renter gets a countdown and 60 seconds to save when it fires. Each notice costs you the renter's setup window (10–60 min) plus the last 10 minutes, refunded to them. Ending a rental without notice counts as a breach: full refund.
Full detail and what each option does to your earnings and reliability score: Host policies.
Step 4: availability and idle-only
With Idle only on (default) your machine is listed only when you have not touched keyboard or mouse for a few minutes and no program uses the GPU. When you come back, no new rentals start, but a running one is not killed; use the policy above to get the PC back, or Pause hosting in the tray before you sit down.
Availability lets you set weekly windows (for example nights and weekends); outside them the machine is hidden.
Step 5: make the listing attractive
Machine → Listing: a real photo of the rig, a one-line tagline, tags, city. Listings with a photo and a committed policy get booked first. Optional extras under Storage: persistent renter workspaces at your price per GB-month, and a disk budget for pre-downloading popular templates so renters see ⚡ ready in seconds.
What runs on your PC
A small open-source agent: one outbound WebSocket to the FairGPU server, nvidia-smi for telemetry, docker run --gpus with dropped capabilities and no host mounts when a rental starts, a tunnel for the renter's traffic through the relay. It never opens ports, reads your files or uses the GPU when nobody rents. Checkpoints and saved environments are uploaded through one-time links; your PC never holds storage credentials.
Electricity
monthly cost ($) = watts ÷ 1000 × hours rented per month × price per kWh
An RTX 4090 rig (≈ 450 W under load), 12 h/day rented at $0.15/kWh: about $24/month. At $0.35/h you earn 360 × 0.35 × 0.9 ≈ $113, net ≈ $89. Cap the GPU power limit at 80 % (NVIDIA App / Afterburner): little speed lost, much less heat and noise.
Payouts
Earnings accrue per minute while a rental runs and show on the dashboard. Request payout once the balance passes the minimum: Stripe (fastest; Stripe handles identity and tax forms), PayPal or bank transfer. Earnings are income in your country; the ledger is downloadable.
Troubleshooting
- Waiting for approval forever: the code lasts 10 minutes; click Try again in the app and approve while logged into the right account.
- Agent offline: is the PC on and logged in? Is the tray icon there? The status window shows the log;
BAD_TOKENmeans the machine was deleted or the token rotated: sign out and in again. - Docker cannot see the GPU:
wsl --update, then restart Docker Desktop; make sure Use the WSL 2 based engine is on and the Windows NVIDIA driver is current. Linux:sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker. - Shows PAUSED although I did not pause: you are not idle, a game is using the GPU, or Docker / NVIDIA checks failed; the machine page shows the reason.