intel iGPU machine learning acceleration

[mich2k]

Hello,
Wondering if it is possible to use this image with intel iGPU/LP iGPU for ML acceleration as described here

I'm asking this since, to my understanding, the -e field only supports cuda

Thanks

ps: I'm running on unraid so is also a bit trick to configure the hwaccel.ml.yml

[martabal]

ps: I'm running on unraid so is also a bit trick to configure the hwaccel.ml.yml

Did you install immich with the docker-compose install method or are you using this AIO image? I'm asking this because enabling GPU acceleration for machine learning with this image has nothing to do with docker-compose or hwaccel.ml.yml on unraid.

If you use this AIO image, currently only CUDA is supported for machine learning, there are issues with openvino, you can follow the progress here #288

[mich2k]

I'm running this AIO image, and now I understood the issue here… I was talking about OpenVINO indeed, ML acceleration is only possible with immich on baremetal then..

Furthermore, I was not updated about the python version requirements and thought was already possible, is a true pity because cpus really struggle with thousands of medias to process

Will follow the issue and hope for a resolution, to my understanding the worst case should be on april (ubuntu24 lts release date)

Thanks for tracking down this issue

[martabal]

Yeah that's something that I really want too.

I created a custom Docker image that compiles Python 3.10 and installs onnxruntime-openvino, but even with that it didn't work. Some users running immich with the official machine learning image reported the same errors I had, I hope everything gets better with 1.17