Step-by-step host setup

Connect your NVIDIA GPU machine to the Edge Compute network. Customers rent your GPUs through our platform; we handle billing, support and security. Plan on about an hour for a fresh machine.

1. Apply to host ~5 min

Hosting is by invitation while we grow the network. Apply below; once approved, we send you a one-time join token that ties your machine to your host account. Treat it like a password.

Want hardware that's ready out of the box? See our Edge Node.

2. Minimum requirements ~1 min

Your machine must meet all of these to be approved.

3. Hardware & BIOS ~10 min

• Above 4G Decoding and Resizable BAR: enabled.
• IOMMU / SVM (AMD) or VT-d (Intel): enabled (needed later for VM isolation).
• Secure Boot: disabled — it blocks the NVIDIA kernel module unless you sign it yourself.
• Restore on AC power loss: Power On, so the machine comes back after an outage.
• PCIe slots: confirm each GPU links at Gen 4/5 and ×8 or wider (we check this after install).
• Fans: set a curve for 24/7 load; GPUs should stay under ~80 °C sustained.
• Two high-power GPUs need a native 12V-2×6 cable each — never daisy-chain adapters.

4. Install Ubuntu & plan storage ~20 min

Option A — USB auto-installer (easiest)

We provide a Ventoy USB stick (or the files to make one) that installs Ubuntu, the driver and our software with no typing. It erases the largest disk in the machine. Boot from USB, pick the Ubuntu image, confirm, and come back in ~30 minutes. Skip to Verify.

Option B — manual install

Download Ubuntu Server 24.04 (amd64), install it with OpenSSH server enabled, then update everything:

sudo apt-get update && sudo apt-get upgrade --with-new-pkgs -y && sudo reboot

Recommended layout — keep customer containers on their own fast partition:

Filesystem   Size     Type   Mounted on
/dev/nvme0n1p1  1G     vfat   /boot/efi
/dev/nvme0n1p2  100G+  ext4   /                 # Ubuntu + host software
/dev/nvme0n1p3  REST   xfs    /var/lib/docker   # customer containers

To set up the Docker partition (replace /dev/<partition>):

sudo mkfs.xfs -f /dev/<partition>
sudo mkdir -p /var/lib/docker
echo "UUID=$(sudo blkid -s UUID -o value /dev/<partition>) /var/lib/docker xfs rw,auto,pquota 0 0" | sudo tee -a /etc/fstab
sudo mount /var/lib/docker && df -h /var/lib/docker

XFS with project quotas (pquota) lets us enforce per-container disk limits.

5. NVIDIA driver ~10 min

Our installer installs the recommended driver automatically if nvidia-smi doesn't work yet, so most hosts can skip this step. If you install it yourself:

sudo ubuntu-drivers install        # picks the recommended (open) driver
sudo reboot
nvidia-smi                         # every GPU listed, no errors

• RTX 50-series (Blackwell) requires the open kernel modules, driver 570 or newer.
• Disable Secure Boot first, or the module won't load.
• Run nvidia-smi -q | grep -i "link" to see each GPU's PCIe generation and width.

6. Network ~10 min

Our installer also adds a firewall so customer containers can't reach your LAN, your router or the host itself.

7. Install the host software ~15 min

Run this on the machine with the join token we sent you:

curl -fsSL https://clearlayer.ca/install.sh | sudo bash -s -- --token join_XXXXXXXX

It installs and configures:

• NVIDIA driver (if missing), Docker and the NVIDIA Container Toolkit (≥ 1.17.8)
• Docker hardening: user-namespace remapping and live-restore
• The sandbox firewall, the security self-check and safe automatic updates
• The edge-agent service, which registers your machine

Options: --name my-node-1 sets the name shown in your dashboard. If the driver was just installed, the script asks you to reboot; the agent starts on its own afterwards.

systemctl status edge-agent          # should be "active (running)"
journalctl -u edge-agent -f           # live log

8. Verify your machine ~10 min

Within a minute the agent connects and runs our security self-check, which starts a container exactly like a customer's and tries to break out of it. Run it yourself any time:

sudo bash /opt/edge-agent/edge-selfcheck.sh

The result is pass, warn or fail. A machine that fails is never shown to customers. We then review the machine (hardware details, PCIe links, a test rental) before approving it.

9. Go live

• Once approved, each GPU appears in the GPU console and can be rented by the minute or reserved.
• Prices are set by us, based on demand; reliability (uptime) is shown to customers and affects how often you're chosen.
• Never interfere with a rented machine — don't reboot it, stop Docker or unplug it while a customer is on it. Use maintenance mode instead.

10. Ongoing maintenance

Automatic updates (nothing to do)

• The NVIDIA Container Toolkit is updated straight away — it only affects new containers.
• Ubuntu security updates install automatically.
• Driver and kernel updates wait: the machine stops taking new rentals, and once the last customer is done it installs the updates and reboots by itself.

Planned maintenance

Put the machine into maintenance mode; current customers keep running, no new ones start:

sudo touch /var/lib/edge-agent/maintenance     # stop taking new rentals
docker ps --filter label=edge.rental_id        # wait until this is empty
# … do your work, reboot, etc. …
sudo rm /var/lib/edge-agent/maintenance        # back in service

11. Troubleshooting

12. Uninstall

Enable maintenance mode and wait for rentals to finish first, then:

sudo systemctl disable --now edge-agent edge-update.timer edge-firewall
sudo rm -rf /opt/edge-agent /etc/edge-agent /var/lib/edge-agent /etc/systemd/system/edge-*
docker ps -aq --filter label=edge.rental_id | xargs -r docker rm -f

Then tell us so we can remove the machine from your account.

Apply to host