Immich- Machine Learning Image Type

I’m on Goldeye now, so the venerable 1080Ti has been swapped for an Arc B50. I’d like to set Immich to use that new card, but I’m gun shy after having to nuke and rebuild immich from scratch after they changed the way storage worked.

Is it as simple as changing the machine learning image type to openVINO in the app edit screen and continuing as usual, or does changing that image setting reset the immich container entirely?

B Series cards are not yet supported on 25.10 as far as i know…

Yea… It’s partially working in jellyfin, but still no AV1. I’m just after whatever I can get until they add support. I should have just scraped up another $150 and gotten the 5060ti before the memorypocalypse hit :confused:

Just looking ahead for when when support is added, will changing that ML image setting nuke my app container? Or is it just like changing cpu and ram limits?

I don’t know enough about immich or the ML container to answer that question.

I also bought a Arc b50 and It seems that just changing it to the Openvino one works fine.

Huh, maybe they worked out the bugs. I wound up returning the b50 for a5060ti, only to dicover that my mobo wont recognize it. Now im just planning to upgrade the whole systAaaaand now ram is infinity dollars.

Computers are great! :upside_down_face::upside_down_face::upside_down_face:

Yeah, homelab upgrades are probably on hold for a lot of people. My test box is cutting edge but with 64gb and not much work to do, it’s gonna stay exactly as it is for the forseeable future.

So yeah, computers are great, as long as someone else is buying.

In your situation I’d be digging into the bios and trolling reddit for answers. The fact that a board can’t see a modern PCIe card is very unusual. Maybe overloading the power needs with the funky 12v cable, who knows. But I’ll just watch my test server hum along with a tiny little one-fan Sparkle card getting all its power from the motherboard and transcoding at light speed. Intel Quicksync is pretty amazing.