RPI5 Server, Intel/Nvidia Gaming PC

[michael-savage]

I recently purchased a RPI5 + M.2 NVME 256gb drive, with the main reason for the purchase being to run Immich 24/7, instead of having the more power hungry PC on.

I'm storing the server, db and thumbs on the RPI'Ss NVME and the library and encoded videos on a NAS.

I have a few questions...

1. I have the PC setup as a machine-learning server, and Immich set up to try that before falling back to the RPI. Is there any way to get the PC to do more work when it's available? It's currently doing some machine learning tasks (every 1/2 from what I've read), but it doesn't seem to be doing the video encoding. It would be really nice to get the PC to do any heavy task when available, and the RPI can just slowly do them when the PC isn't available. Also, rather than split the tasks 1/2, if the PC is available it could do ALL the work. Video encoding jobs could stream to the PC and stream back the encoded data?

2. We have 4 libraries. Mine, my wife's and my daughters, along with an "admin" library that contains a mix of all from pre-Immich days. Admin shares to my wife and me, and we also share to each other. My daughter has her own library shared to us, but doesn't have anything shared with her (currently). As far as I'm aware, all the detected people need to be managed separately. That's doable, but I'd prefer a central repository of all people that all all libraries draw off. It makes sense to be that way as person is still the same person no matter which library they're in.

3. I forget what the last point is. I may as well use it to say how freaking awesome Immich is, and I plan to contribute to it as much as possible shortly.

[michael-savage]

I should mention that I think the best way currently is to get the PC to do the initial load then move all the database, thumbs to the RPI. The RPI is agonisingly slow for the initial population but should be fine to keep up with incremental uploads.

[mertalev]

While not as easy as remote machine learning, it's possible to offload other tasks to the other server as well. There's a helpful guide about it here.

Doing the initial import on the other machine is also a viable option.

[michael-savage]

Thanks for the link! I hadn't seen that. I also wonder if Immich would handle sharing a db with multiple server instances. So the db would reside on the RPI, but the PC would have all the other servers, hw accel, etc running when it's up, but pointing to the RPI db.

[mertalev]

Yes, that would work fine. There is a caveat that the DB has a big effect on overall performance, particularly for daily usage when scrolling the timeline, etc. An SSD makes a big difference here if any of the other devices have one. An external SSD for the Raspberry Pi is also viable.

Oh, I didn't see you have an NVMe on the Pi. That seems like a good setup.