📘 VastAI GPU Host Guide Professional GPU Rental Setup by AG Tech Sec

Welcome to VastAI-GPU-Host-Guide, a full, production-grade blueprint for turning an Ubuntu server into a GPU rental node on Vast.ai — the world’s largest peer-to-peer GPU marketplace.

This project documents every step performed on the HP Server (RTX 3060) and is designed so anyone can set up their own GPU host securely, reliably, and profitably.

Built for:

🧠 Engineers expanding their infrastructure skills

💵 People who want passive income from unused GPUs

🔐 Professionals interested in secure, self-managed hosting

🖥️ Home-lab enthusiasts running NVIDIA hardware

🌍 Overview

This guide documents the entire end-to-end process of:

Preparing Ubuntu for GPU hosting

Fixing NVIDIA runtime issues

Installing Vast.ai daemon

Passing verification

Listing your GPU on the marketplace

Enabling auto-recovery & stability

Handling GPU swaps (1060 → 3060)

Maintaining your host over time

Troubleshooting advanced runtime problems

Improving system security

The project includes screenshots, scripts, and deep troubleshooting, based on real-world installation and repair logs.

🧩 Tech Stack Component Purpose Ubuntu 22.04 / 24.04 Base OS NVIDIA Driver 535 GPU driver for CUDA & Vast NVIDIA Container Toolkit Exposes GPU → Docker containers Docker + NVIDIA Runtime Runs GPU workloads for renters Vast.ai Kaalia Host Daemon Marketplace connection Systemd Autostart Ensure reliability Port Forwarding (16384–32768) Required for renter connections 🚀 Key Features

📦 Full GPU passthrough inside containers

🛠️ Self-healing configuration and runtime repair

🔁 Auto-pull of PyTorch container

🧪 GPU verification scripts

🔒 Security hardening included

🔧 Correct NVIDIA CDI injection (most common Vast failures)

🔌 Supports GPU upgrades & swaps

💸 Optimised pricing templates

💾 Supports NVMe upgrades & multi-GPU expansion

📸 Screenshots (Placeholders)

Replace with your screenshots from /IMAGES.

IMAGES/install_success.png

IMAGES/gpu_3060.png

IMAGES/docker_runtime_fixed.png

IMAGES/cdi_generated.png

IMAGES/vastai_listing.png

🔧 Folder Structure VastAI-GPU-Host-Guide/ │ ├── README.md ├── INSTALLATION.md ├── TROUBLESHOOTING.md ├── GPU_SWAP.md ├── MAINTENANCE.md ├── SECURITY.md │ ├── scripts/ │ ├── verify_gpu.sh │ └── auto_restart_docker.sh │ └── IMAGES/ └── (all your screenshots)

🧠 Why This Project Matters

Building this project gives strong, practical experience in:

Linux system administration

Docker orchestration

GPU runtime debugging

Server hosting architecture

Secure self-hosting

Networking (port forwarding, static IPs, firewall rules)

Troubleshooting real vendor bugs (NVIDIA runtime, CDI)

Cloud platform operations

Market-based GPU pricing

This is high-value knowledge for:

AI hosting

DevOps

Cloud engineering

Cybersecurity

Home lab design

Freelance system architecture

And yes — it can generate profit if a good GPU is listed (e.g., RTX 3060, 3070, 4090, A2000).

👤 Author

Adam Gwozdz (AG Tech Sec) GitHub: https://github.com/AG-Sec4

Website: https://agtechsec.com