Skip to main content
Inferix

Complete Host Setup Guide

Comprehensive walkthrough for GPU providers: from account setup through optimization and verification.

1. Prerequisites & Hosting Agreement

Convert Your Account to a Hosting Account

Before you can host machines, you must agree to our hosting agreement and convert your account to a hosting account. This is a one-way conversion that changes your account permissions and access.

Important

Once converted to a hosting account, you will have different permissions than a regular user account. You'll gain access to:

  • ✓ Host-only dashboard and machine management
  • ✓ Provider earnings tracking and payouts
  • ✓ Dedicated host support channels
  • ✓ Host community forums

To convert your account:

  1. Go to Settings → Account Type
  2. Review and accept the Hosting Agreement (required)
  3. Click Convert to Hosting Account
  4. Your account will be converted immediately

Once converted, you can access our host community on Discord. Link your hosting account to gain access to our #host-support and #host-discussion channels for help and to connect with other providers.

2. System Requirements

Your machine must meet minimum requirements, but the ideal machine will exceed these guidelines, as some workloads require more resources.

Operating System

  • Ubuntu 18.04 or newer (required)
  • Ubuntu 22.04 recommended (latest, best hardware support)
  • Ubuntu 20.04 acceptable but may have slower performance on some workloads
  • Server edition preferred over desktop

Hardware Requirements

ComponentMinimumNotes
GPU10-series Nvidia or newerAll GPUs must be same type; older Nvidia not recommended
CPU1 core per GPU (2 hyperthreads)Must support AVX instruction set
RAM4 GB per GPU8+ GB per GPU recommended for better workload support
Storage128 GB SSD per GPUNVMe required for XFS partition and optimal performance
PCIe1x per 2.5 TFLOPS GPUPCIe 3.0 x16 recommended per GPU
Network10 Mbps per machine100+ Mbps recommended; stable uptime ≥99%

⚠️ Important: Dedicated Machines Only

Your machine should be dedicated to hosting. Don't host from a client account or run competing workloads. If your machine goes offline or crashes while a client is paying, it permanently damages your reliability rating.

3. Install Ubuntu OS

Download Ubuntu ISO

Download the appropriate Ubuntu ISO for 64-bit PC (AMD64):

  • Desktop or Server edition: Both work; Server recommended for headless setups
  • Ubuntu 22.04 LTS recommended (latest stable with long-term support)
  • Download from: https://ubuntu.com/download/server

Disk Partitioning Strategy

Critical: Our software requires an XFS partition for optimal performance. By default, this isn't created during Ubuntu install, so we need to set it up manually.

Option A: Single Drive Setup (Most Common)

If you have a single SSD for both OS and data:

  1. 1. On "Installation Type" step, select "Something Else"
  2. 2. Click button on right to create new partition table (if needed)
  3. 3. Create partitions in order:
EFI - ~512 MB - /boot/efi
ext4 - 32 GB - /
(unallocated) - Remaining - (for XFS)

4. Leave rest of drive unallocated. Installer will create XFS partition from unallocated space.

Important Notes

  • Don't create a swap partition — causes major performance issues with Docker
  • ✓ Create EFI partition if your system uses UEFI
  • ✓ Leave plenty of unallocated space for XFS (at least 50% of drive)

4. Install GPU Driver & Docker

$sudo apt update && sudo apt upgrade -y
sudo apt install nvidia-driver-550 -y
nvidia-smi -q

sudo apt install docker.io nvidia-docker2 -y
docker run --rm --gpus all nvidia/cuda:11.0-runtime nvidia-smi

Ubuntu 22.04 Only: Fix Cgroup Issue

Edit /etc/docker/daemon.json:

${
  "exec-opts": ["native.cgroupdriver=cgroupfs"],
  "runtimes": {
    "nvidia": {
      "path": "nvidia-container-runtime"
    }
  }
}
$sudo systemctl restart docker
docker info | grep "Cgroup Driver"
# Should output: Cgroup Driver: cgroupfs

5. Install Platform Manager Software

Once your Ubuntu OS is fully installed and drivers are working, you'll install our platform manager software.

Get Your Installation Command

  1. Log in to your provider dashboard
  2. Go to Machines → Add Machine → Installation Command
  3. Copy the personalized installation command (valid for 1 hour)

Run the Installation

$wget https://api.platform.dev/install -O install
sudo python3 install <YOUR_TOKEN_HERE> --interactive
history -d $((HISTCMD-1))

Installation flags:

  • --interactive: Recommended (prompts you through decisions)
  • --no-docker: Use if daemon update fails
  • --raidgpt: If using RAID with GPT partition table

6. Configure Networking

Open Port Range

Port Requirements:

  • • Minimum: 1 per GPU
  • • Recommended: 100+ per GPU
  • • Example: 4 GPUs = 400 ports (16000-16399)
  • • Both TCP and UDP must be open

Router Configuration

  1. Log into your router admin panel
  2. Find Port Forwarding section
  3. Create mapping: 16000-16399 → your machine's local IP
  4. Save and apply changes

Test Your Ports

$nc -l -p 16000

Visit portchecker.co and verify port 16000 with your public IP.

Tell Daemon About Ports

$echo -n "16000-16399" | sudo tee /var/lib/vastai_kaalia/host_port_range

Daemon will update within 5 minutes showing new port range.

7. List Your Machine

Once your machine is registered and online, you can list it for rental immediately.

Via Dashboard

  1. Go to Machines page
  2. Find your newly registered machine (wait 1-5 minutes)
  3. Click List button
  4. Set hourly price (e.g., $0.65 per GPU)
  5. Set availability end date
  6. Click Confirm Listing

⚠️ During Active Rental:

  • ✅ Keep machine online and stable
  • ✅ Don't interfere with the machine
  • ❌ Don't restart daemon or reboot
  • ❌ Don't upgrade/downgrade components
  • ❌ Don't run competing workloads

8. Machine Verification

Verification Status

When first listed, your machine has Unverified status:

  • ✓ Machine is searchable
  • ✗ Won't appear in default search filters
  • ✗ Won't show in sorted-by-reliability lists

How to Get Verified (Fast)

  • ✓ Maintain ≥99% uptime
  • ✓ Pass performance benchmarks (run self-test)
  • ✓ Reliably complete jobs without crashes
  • ✓ Keep stable network connection
  • ✓ Don't change components during verification
$vastai self-test machine <MACHINE_ID>

Timeline: 24-48 hours to verified (if all criteria met)

9. Optimization & Maintenance

Disable Automatic Driver Updates

$sudo apt-mark hold nvidia-driver-*

Set Minimum Bid Price

$vastai set machine <MACHINE_ID> min_bid 0.50

Customize APT Mirrors

If your machine is far from US-based mirrors:

$echo 'mirror.yandex.ru' | sudo tee /var/lib/vastai_kaalia/apt-select-out

10. Troubleshooting & Support

ProblemSolution
Machine doesn't appearWait 1-5 min, check: sudo journalctl -u vastai_kaalia -f
nvidia-smi not foundInstall: sudo apt install nvidia-driver-550
Docker won't startInstall Nvidia runtime: sudo apt install nvidia-docker2
Ports unreachableVerify port forwarding rules in router admin panel
Machine deverifiedRun self-test to diagnose: vastai self-test machine <ID>

Get Help

  • Discord: #host-support and #host-discussion (link your hosting account first)
  • Dashboard: Click support chat widget (bottom right)
  • Email: support@platform.dev

🎉 Ready to list your machine?

You now have everything you need to become a provider and start earning. Follow the steps in this guide, and you'll be live with your first machine in 2-4 hours.