Jellyfin GPU Passthrough on Proxmox
Not really a PROJECT, but part of a big reorganization of the homelab. I moved the whole media stack to “Celebrimbor” — an HP Pavilion 15et something something with an MX550 GPU. Now it runs Jellyfin with hardware transcoding via GPU passthrough to a Proxmox LXC. It way OP for a media server, but the MX550 supports NVENC/NVDEC, which means proper hardware transcoding instead of CPU-only. It’s not much for AI inference. 2GB VRAM is too small.
| Component | Details |
|---|---|
| Machine | Pavilion 15t-e300 (nk-celebrimbor) |
| GPU | NVIDIA MX550 (2GB VRAM) |
| Setup | Proxmox LXC with GPU passthrough |
The machine was previously an AI inference node but got decommissioned from the AI fleet. All LLM inference is now on Legolas (Mac Studio). The MX550 is dedicated entirely to Jellyfin transcoding now.
Full Servarr stack runs alongside Jellyfin — Radarr, Sonarr, Prowlarr, qBittorrent behind Gluetun VPN. oops I mean iso downloaders. Multiple simultaneous streams transcode fine with the MX550.