immich
immich copied to clipboard
immich-app
[QUESTION] Illegal instruction (core dumped)
Describe the bug A clear and concise description of what the bug is.
docker-immich_microservices-1 |
docker-immich_microservices-1 | > [email protected] prebuild
docker-immich_microservices-1 | > rimraf dist
docker-immich_microservices-1 |
docker-immich_microservices-1 |
docker-immich_microservices-1 | > [email protected] build
docker-immich_microservices-1 | > nest build
docker-immich_microservices-1 |
docker-immich_microservices-1 |
docker-immich_microservices-1 | > [email protected] start:prod
docker-immich_microservices-1 | > node dist/main
docker-immich_microservices-1 |
docker-immich_microservices-1 | Illegal instruction (core dumped)
docker-immich_microservices-1 exited with code 132
I am running this on a machine with a low-power Celeron CPU that does not have AVX instruction support, so that seems like a likely cause.
I assume this is the ML stuff? Is it possible to disable AVX instructions for these? Could I pass through the intel iGPU to do this on instead?
Ah, should've read the README to the end...
It'd be nice if immich-microservices could be run on non-AVX CPUs and/or iGPUs instead though.
Hi, unfortunately, it is TensorFlow that uses the AVX instruction. The only workaround is to build it from the source and disable it, the distributed node.js library that comes with Immich has it enabled. :/
I see.
Can TF be made to use the iGPU? Perhaps it doesn't use AVX when using a GPU.
I have an experimental dockerfile that uses GPU but haven't actually gotten a chance to fully work on that to incorporate into the code. I will keep this issue opened and updated when I have more info
Me too I have a Synology NAS with INTEL Celeron J4125, I really would like to try immich-microservices on iGPU... If you need help on some test, I'm here...
Which nodejs package has the AVX dependency? Is there a non-avx version of the pkg? Im looking into building a iGPU multi-stage dockerfile to get this going!
This might help to build tfjs-node without avx: https://peterbabic.dev/blog/install-tfjs-node-from-source/
Is there any update on this? I have a NAS with Intel Celeron J3455 that gives the same error, and now I can't run the machine learning container, causing more errors.
