Immich NAS + GPU Processing Setup

I need Immich running on my QNAP TS-216G so that every original image and video, the PostgreSQL database, and all generated thumbnails live permanently on the NAS itself. Heavy tasks—face recognition, video transcoding, metadata extraction—should be pushed to my Windows-based desktop with an NVIDIA GPU and handled in real time over the network. Once processing is done, the NAS must still serve and display content without relying on the PC being online.

Here’s what I’m after:

• Install and configure Immich (Docker or native, your choice) on the QNAP with all persistent volumes mapped to NAS storage.
• Set up a clean mechanism for off-loading compute jobs to the Windows PC: this could be a companion Immich micro-service stack, Docker over WSL2, or any other approach you recommend. I’m open to tools as long as the GPU is fully utilised.
• Ensure a seamless hand-off so that uploads, indexing and on-the-fly edits happen automatically, yet library access never breaks when the PC is switched off.
• Harden the installation—user permissions, secure networking between NAS and PC—and document every command, config file and port used so I can reproduce or update the stack later.

Deliverables will be:
1. A working Immich instance on the TS-216G storing everything locally.
2. Verified GPU-accelerated processing from the Windows machine, demonstrated with both images and videos.
3. Clear step-by-step notes (Markdown or PDF) covering setup and future maintenance.

If you’ve tackled similar NAS + GPU split-processing workflows before, tell me how you did it and the performance you achieved. I’m ready to start as soon as you can outline the strategy.