Skip to content

Rent out your GPU

Any Linux machine with a supported NVIDIA GPU can join the marketplace. The Gysga agent runs renters’ containers, reports your GPUs and keeps an outgoing tunnel to the platform — you don’t open any ports.

  • OS: Ubuntu 22.04 or 24.04, x86_64, with systemd.
  • GPU: one or more supported NVIDIA GPUs of the same model in one machine (see below).
  • NVIDIA driver 570 or newer (the current templates need CUDA 12.8–12.9; older drivers get fewer renters because templates with a higher CUDA requirement skip the machine).
  • Docker and the NVIDIA Container Toolkit (docker run --gpus all nvidia/cuda:12.8.1-base-ubuntu24.04 nvidia-smi must work).
  • Disk: 100+ GB free on the partition with /var/lib/gysga (images, models and renters’ /workspace live there). More disk = more templates and larger models fit.
  • RAM: at least 16 GB per GPU is recommended. CPU cores and 90% of RAM are shared between renters in proportion to their GPUs.
  • Network: a stable connection; 100+ Mbit/s recommended for downloading models. No public IP or open ports needed.
  • The machine should be dedicated: renters get the GPUs exclusively while their servers run.

RTX 3060 (12 GB), 3080 Ti, 3090, 4060 Ti 16GB, 4070 Ti SUPER, 4080 / SUPER, 4090, 5070 Ti, 5080, 5090; RTX A4000, A5000, A6000, RTX 6000 Ada, RTX PRO 6000 Blackwell; A10, A40, L4, L40, L40S, A100 40/80GB, H100, H200, B200. Laptop GPUs and cards with less memory than listed are not accepted. Renters pay the fixed price of your GPU model.

  1. In the dashboard open Rent out your GPU → Add machine. You get a one-time install command; its token is valid for 24 hours and registers one machine.
  2. Run it on the machine:
Terminal window
curl -fsSL https://get.gysga.com | sudo sh -s -- <token>

The installer:

  • checks the OS, the NVIDIA driver, Docker and the NVIDIA Container Toolkit;
  • downloads the gysga-agent binary;
  • writes its configuration to /etc/gysga/ (the machine’s credentials are stored in /etc/gysga/machine.json, readable by root only);
  • installs and starts the gysga-agent systemd service.

On start the agent:

  • creates the Docker network gysga (10.231.0.0/16, no traffic between containers, no IPv6);
  • adds firewall chains GYSGA-FWD and GYSGA-IN (iptables) that block containers from reaching your LAN, private address ranges, cloud metadata, the host itself and outgoing SMTP;
  • stores data in /var/lib/gysga (models/ — model cache, instances/ — renters’ workspaces).

Within a minute the machine appears in the dashboard with its GPUs. When all GPUs are supported it becomes active and starts receiving renters.

You earn a share of the rental price for every second a renter’s server runs on your machine:

Level Share Requirement
Bronze 65% start level
Silver 75% 95% uptime, machines active for 7+ days
Gold 85% 99% uptime, machines active for 30+ days

Uptime is the share of time your machines are online. The level is recalculated automatically; a server keeps the share that applied when it was launched. Higher-level and more reliable machines are offered to renters first.

Example: one RTX 4090 ($0.40/h) rented 16 hours a day earns about $125/month at Bronze and $163/month at Gold.

  • Earnings are on hold for 7 days (to cover disputes), then become available.
  • Payouts are made in USDT on TRON (TRC20). Set your address in Rent out your GPU → Payouts.
  • Minimum payout: $10. Request any available amount; the team sends it and the transaction hash appears in the payout history.
  • After changing the payout address, payouts are paused for 48 hours (protection against account takeover).
  • Stop renting temporarily: sudo systemctl stop gysga-agent. The machine goes offline; running renters are paused (not charged) and your reliability drops, so avoid stopping while servers run.
  • Remove a machine in the dashboard. It’s only possible when no renters’ servers are left on it; the agent’s credentials are revoked.
  • Uninstall:
    Terminal window
    sudo systemctl disable --now gysga-agent
    sudo docker rm -f $(sudo docker ps -aq --filter label=gysga.instance) 2>/dev/null
    sudo docker network rm gysga
    sudo rm -rf /etc/gysga /var/lib/gysga
Terminal window
journalctl -u gysga-agent -f # live logs
sudo systemctl restart gysga-agent
nvidia-smi # GPUs visible?
docker run --rm --gpus all nvidia/cuda:12.8.1-base-ubuntu24.04 nvidia-smi
Symptom Cause
Machine stays pending A GPU is not supported, or GPUs of different models are installed.
not registered: set GYSGA_INSTALL_TOKEN The token was not passed; create a new command in the dashboard.
Install token is invalid, used or expired Tokens are single-use and valid for 24 h; create a new one.
Renters’ servers stay in pulling Slow download or low disk space; check df -h /var/lib/gysga.
  • No inbound ports. The agent only makes outgoing HTTPS connections to agent.gysga.com.
  • Containers are unprivileged: no --privileged, dropped capabilities (NET_RAW, MKNOD, AUDIT_WRITE, SYS_CHROOT), no-new-privileges, limits on CPU, memory (no swap) and processes.
  • No access to your network: the firewall chains block private ranges (10/8, 172.16/12, 192.168/16, 100.64/10), link-local and multicast addresses, the host itself and SMTP.
  • Renters’ data is deleted from your disk when they destroy their server.

Keep the machine updated and don’t run other important services on it. Renters run arbitrary code in containers; container isolation is strong but not a virtual machine.