Chris wrote:
Got it — this is an automatic acknowledgment from Chris, your AI support representative. Your post is in the support queue, and we’ll get to it as soon as we can. Legend members receive priority replies. If you’d like faster support while helping fund the project’s goals and continued development, consider becoming a Legend backer. <!-- chris-support-queue-ack:v1 --
Chris wrote:
Hi markof — you're right on both counts, and the first one was our mistake. Immich 3.1.0 vs 3.2.1. SparkBox pins the exact Immich version and only a SparkBox update moves it (a hand-edited compose file gets put back by the next update). Our nightly updater is supposed to carry Immich forward on its own, but it was looking for Immich's releases in the wrong place, so it never saw 3.2.0, 3.2.1 or 3.2.2. That is fixed, and the move to Immich 3.2.2 is in the next SparkBox release — we tested it on our own bench as a real upgrade (a library on 3.1.0 with an admin and photos → 3.2.2: all migrations ran, photos and login intact). A library that is already on 3.2.1 opens fine on 3.2.2; it is a small patch on top of 3.2.1. Legend boxes get the new release the day it ships; free boxes follow about a week later on the stable channel. If you would rather not wait, the supported way to move Immich ahead today is a per-app override, which survives updates: 1. sudo mkdir -p /opt/sparkbox/state/overrides 2. Create /opt/sparkbox/state/overrides/immich.yml containing (two spaces per indent level): - line 1: services: - line 2: immich-server: - line 3: image: ghcr.io/immich-app/immich-server:v3.2.2 - line 4: immich-ml: - line 5: image: ghcr.io/immich-app/immich-machine-learning:v3.2.2 3. sudo sparkbox up immich Once the SparkBox update that carries 3.2.2 has landed on your box, delete that file (or at least the two image lines) so later Immich updates flow again — an override pins whatever tag it names. CUDA image for the machine-learning container. There is no built-in switch for that yet — I've logged it for Tom as a feature request. Today the GPU is wired into Immich's server (video transcoding) when hardware acceleration is set to nvidia and the NVIDIA container runtime is registered with Docker, but the ML container stays on the CPU image. You can do the CUDA part yourself with the same override file, and it survives updates too: - line 4: immich-ml: - line 5: image: ghcr.io/immich-app/immich-machine-learning:v3.2.2-cuda - line 6: runtime: nvidia - line 7: environment: - line 8: - NVIDIAVISIBLEDEVICES=all - line 9: - NVIDIADRIVERCAPABILITIES=compute,utility Then sudo sparkbox up immich. Before you add that block, run sudo sparkbox doctor and look for the line saying the Nvidia container runtime is registered — if it is not, runtime: nvidia will stop the ML container from starting until the NVIDIA Container Toolkit is installed and registered with Docker. Immich's CUDA image also wants a fairly recent NVIDIA driver (their docs ask for 545 or newer) and a card with compute capability 5.2 or higher. If your Immich server should transcode video on the GPU as well, sudo sparkbox set-env JELLYFINHWACCEL nvidia followed by sudo sparkbox up turns that on for Immich's server (and Jellyfin) without any override. If the migration itself throws anything after the version lines up, paste the first error line from sudo sparkbox logs immich-server and I'll take it from there.