Tdarr: NVIDIA GPU Support

[ltsch]

Tdarr: NVIDIA GPU Support

Summary

Adds NVIDIA GPU hardware acceleration support to Tdarr LXC containers.

Important

This PR prioritizes the native var_gpu and setup_hwaccel for Intel/AMD iGPU support. This PR adds NVIDIA support for /dev/nvidia* devices in addition to the standard /dev/dri interface.

Note

This PR modifies two files: ct/tdarr.sh and install/tdarr-install.sh.

NVIDIA Requirements

GPU Type	Device Interface	Handled By
Intel/AMD iGPU	/dev/dri (render nodes)	Native var_gpu + setup_hwaccel
NVIDIA	/dev/nvidia* (proprietary)	This PR (explicit passthrough + version-matched drivers)

NVIDIA GPU passthrough to LXC containers requires:

1. Explicit device passthrough — /dev/nvidia* devices are not part of /dev/dri
2. Exact version matching — Container userspace libraries must match host kernel driver

Without this, applications attempting to use the GPU (e.g., handbrake, ffmpeg) will fail with a Driver/library version mismatch error when attempting to initialize the NVML or CUDA interface.

Core Libraries & Transcoding

The following libraries are matched to the host version to enable hardware acceleration:

Library	Role	Enables
libcuda1	CUDA Base	Communication with GPU hardware
libnvcuvid1	NVDEC	Hardware Decoding in FFmpeg/Handbrake
libnvidia-encode1	NVENC	Hardware Encoding in FFmpeg/Handbrake
libnvidia-ml1	NVML	Device monitoring and temperature/load reporting

Service Management

The PR implements robust systemd service management to ensure reliability and ease of use:

• tdarr-server.service: Manages the main Tdarr Server. Includes ExecStartPre to automatically run the Tdarr_Updater before every start.
• tdarr-node.service: Manages the Tdarr Node. Configured with Requires=tdarr-server.service to ensure the correct startup sequence.
• Auto-Enable: Both services are enabled to start automatically on container boot, replacing the manual or cron-based startup requirement.
• Configure a server or node: tdarr-node or tdarr-server services can be enabled or disabled per container.

---

Changes

ct/tdarr.sh (Host-Side)

Change	Purpose
NVIDIA detection	Identifies driver version to offer optional pinning
Driver pinning	Optional apt-mark hold on host to prevent breaking updates
Help/Abort flow	Prevents "half-broken" installs if drivers are missing on host
Uses community packages	Delegates passthrough logic to build.func natives

install/tdarr-install.sh (Container-Side)

Change	Purpose
Autonomous detection	Reads /proc/driver/nvidia/version (shared with host)
Version matching	Installs libcuda1 etc. pinned to host version
Verification tools	Installs nvidia-smi for status monitoring
Systemd Services	Configures robust tdarr-server and tdarr-node services
Intel VA-API	Optional intel-media-va-driver-non-free for multi-GPU support

---

Architecture

┌─ ct/tdarr.sh (Proxmox Host) ─────────────────────────────┐
│  1. Detect NVIDIA driver / Pin host drivers (Optional)   │
│  2. Delegate passthrough to build.func (var_gpu="yes")   │
└────────────────────────────────────────────────────────────┘
                           │
                           ▼
┌─ install/tdarr-install.sh (Inside Container) ────────────┐
│  1. Detect driver version directly via /proc             │
│  2. Pin NVIDIA CUDA repo + Install matched packages      │
│  3. Install nvidia-smi for user verification             │
└────────────────────────────────────────────────────────────┘

---

Testing

• Verified on Tdarr built on Debian 13 (Trixie) with NVIDIA host driver 590.48
• Verified on Proxmox 8.4.16/368e3c45c15b895c (running kernel: 6.8.12-17-pve)
• Confirmed apt correctly prioritizes pinned NVIDIA repo
• Confirmed services start cleanly

---

Files Changed

• ct/tdarr.sh — NVIDIA detection and device passthrough
• install/tdarr-install.sh — Userspace package installation

[MickLesk]

To be honest, I don't understand the point of the PR. What exactly needs to be fixed ONLY in tdarr?

We have a function in build.func that fits the GPU and installs the necessary drivers (if possible) in tools.func.

[ltsch]

I didn't want to mess with the core build.func used by many projects. Instead I focused on addressing this issue for Tdarr due to my own frustration with the community script not working correctly (container driver version did not align to host driver). If the efforts should be focused on build.func to improve Nvidia driver handling then we can close this PR.