Run Ollama with NVIDIA GPU Acceleration on Proxmox LXC (2026)

Run Ollama with full NVIDIA GPU acceleration inside a Proxmox LXC — no Docker required. This guide walks through every step: blacklisting nouveau, installing NVIDIA drivers on both the host and container, creating a Debian 13 LXC with GPU passthrough, and installing Ollama manually (no install script). We also set up a systemd service, bind Ollama to the network for cross-container access, and verify GPU usage multiple ways. Then we go beyond the basics and productionize it: Pi-hole DNS record, Caddy reverse proxy with Cloudflare SSL, Proxmox Backup Server job with .pxarexclude to skip model files, and Uptime Kuma monitoring for both the API and container. Part 1 of the Paperless-ngx on Proxmox series. Ollama serves as the local LLM backend for Paperless-GPT in Part 2. Written guide with copy-paste commands: https://localhake.com/content/ollama-nvidia-gpu-proxmox-lxc Specs used in this video: - Proxmox VE 9.1 - Debian 13 (Trixie) LXC - NVIDIA Driver 595.58.03 - Ollama (manual install from tarball) - RTX 3060 12GB (GPU passthrough) Related videos: - Pi-hole & Unbound on Proxmox LXC - Caddy with Cloudflare SSL on Proxmox LXC - Proxmox Backup Server Setup Guide - Uptime Kuma: Install & First Monitors - Paperless-ngx on Proxmox LXC (Part 1) - Paperless-GPT on Proxmox LXC (Part 2) 0:00 Intro 1:16 Written Guide & Copy Feature 2:15 Blacklist Nouveau Drivers 3:13 Rebuild initramfs & Reboot 3:57 Install NVIDIA Driver on Host 5:34 Create the Ollama LXC 7:15 GPU Passthrough Configuration 8:30 Install NVIDIA Userspace in LXC 9:55 Install Ollama (Manual Method) 10:41 Systemd Service & Network Binding 12:34 Pull a Model & First Test 13:16 Verify GPU is Being Used 14:54 Pi-hole DNS Record 16:19 Caddy Reverse Proxy with SSL 17:26 PBS Backup with .pxarexclude 19:38 Uptime Kuma Monitoring 20:59 Next Steps