Whatever machine is hosting the Immich server will be doing all the processing

By default yes but it's possible to do remote ML, cf https://immich.app/docs/guides/remote-machine-learning so my "ideal" setup would be all on RPi5 BUT initial ML tasks (because it's going to be > 100GB of photos at first) either directly on my PC with GPU or remotely (cf guide).

I don't actually know BUT it would be amazing to start with a ranking, e.g. https://www.npr.org/2024/07/16/g-s1-9554/best-games-2024-picked-by-npr-staff , and see if 100% of those via e.g. ProtonDB API or even manually (but not ideal to stay up to date), are above a quality threshold, say Gold or Platinum (or obviously native) then tell everyone about it.

Because... if it's true that 100% of those (anti-cheat excluded) do work then even I, a gamer who runs exclusively Linux on PC and handeld, would be assuming maybe 80%.

Does it mean I would do the first import on desktop, including ML, then I could switch the resulting data (via e.g Docker) to a RPi5 if I get <100 new photos per day?

Which... is "funny" because even though it is a genuine arm race where 2 powerful nations are competing... it's a pointless one.

Sure, we do get slightly better STT, TTS, some "generation" of "stuff", as in human sounding text use for spam and scam, images and now videos without attribution, but the actual hard stuff? Not a lot of real change there.

Anyway, interesting to see how the chips war unfold. For now despite the grand claim though from both :

• US with software and models for AI (Claude, OpenAI, etc driven by VC backed funding looking for THE next big thing, which does NOT materialize) ) and hardware, mostly NVIDIA (so happy to sell shovels for the current gold rush) or
• China with "cheap" to train large models (DeepSeek) and hardware (SMIC, RISC based chips) to "catch-up" without any large production batch with any comparable yield

neither have produced anything genuinely positive IMHO.

NVIDIA-SMI 535.183.01 on NVIDIA GeForce RTX 2080 Ti, been playing (and working) with it on Debian (and Ubuntu) with that setup for years now and... pretty much 0 problem.

It's only when I tinker with CUDA for ML tinkering that I might spend 1h re-installing the right driver to match version, or containers, but otherwise, as daily driver, pretty much flawless experience.

I believe from time to time I get a glitch on Plasma when PC comes back from hibernation but that's solved in 1s.

I really need to take the time to

• install (or even update) my local instance
• try a batch import (any hint?) from iOS (iirc the app did work well, just takes a while the first time)
• setup the ML part (on a 2080ti, so that also will probably take time initially)

Since ProtonDB (and obviously Proton itself, Wine with its own WineApp DB, SteamOS) there is an easy way to check if your favorite games do work. That being said I understand that people are afraid. They might think "OK... well Elden Ring works but what about the DLC, or upcoming Elden Ring Nightreign?" and believe, probably rightfully so to be honest, that because Windows is still the most popular OS for gaming on PC and that game publishers are economically rational actors, more testing and fixes will be done against that target platform.

So... 100% is a ridiculous coverage because it's impractical but IMHO they are not that silly to "want" it. It's just a simpler way to say they are scared and do not want to bother. They would rather follow the crowd than take a risk themselves and be trail blazers.

All that being said now that ProtonDB exists and Valve is actively radically improving support via Proton, that gamers see in the wild SteamDecks popping up literally around them, in flights, airports, waiting rooms, etc they just can not ignore the fact that support is improving enough to have fun. Mentality will change but it takes time and Microsoft is fighting back because despite having Azure as their dollar printing asset, they are just hooked on bundling.

I didn't downvote but I bet I understand people who did, as this comment does NOT address OP concern. They just add yet another alternative to verify without explaining how to actually do so, i.e. they make the problem worst rather than help, IMHO.

Since you ask, here are my thoughts https://fabien.benetou.fr/Content/SelfHostingArtificialIntelligence with numerous examples. To clarify your points :

• rely on open-source repository where the code is auditable, hopefully audited, and try offline
• see previous point
• LLMs don't "analyze" anything, they just spit out human looking text

To clarify on the first point, as the other 2 unfold from there, such project would instantly lose credibility if they were to sneak in telemetry. Some FLOSS projects tried that in the past and it always led to uproars, reverts and often forks of the exact same codebase but without telemetry.

I'm sure you do, you little scum! /s

And it should, unfortunately it's not. Maybe right to repair and other laws will, hopefully, change that but for now, it's bundling, part pairing and locks all the way down.

Specially the moment you open the browser

I'd be curious, did you profile if it's for all pages or only some? I'd expect e.g. Facebook or Instagram to be more demanding than Lemmy or ProtonMail but to be honest I have no idea.