Taasz/Woof

Sounds like an utterly terrible idea.

Those are the trackers from websites you’re visiting, not the browser.

Sounds like you want something more custom built if you want to wake up hosts when needed to access the media on them, that’s not a very common setup. Maybe just a plain old set of SMB shares with the images on them for something simple! I suppose you’d want some kind of index kept on an always-on host to be able to figure out which host to wake up.

Distributed storage with all the hosts online 24/7 is more common, and Immich could certainly handle that just fine as your storage backend would be doing the work of distributed storage (Ceph or similar), and Immich doesn’t even need to be aware of it.

Immich runs great on my server with an i5-7500 CPU and uses about 800MB of RAM, with around 3-4GB while actively running the ML models on new uploads using OpenVINO/Quicksync hardware acceleration on the iGPU (RAM usage there depends on which models you’ve picked). I would consider that to be pretty old and slow hardware.

There are definitely other projects for slower hardware out there that are just simple photo galleries. Or an app like FolderSync can backup/upload your photos to any type of file server if you don’t need the web gallery.

Immich came out in '22 and I don’t think makes much use of LLMs, at least it doesn’t feel like it does.

For resource usage did you disable all the search, face recognition, and transcoding modules? Those will have heavy resource usage on CPU if you didn’t set up hardware acceleration for them.

I find it very fast overall and the heavier resource usage is only when photos are initially added.