Hardware‐Accelerated Machine Learning

Hardware-Accelerated Machine Learning

Note

This guide is based on the quadlet files in main.

Warning

As I don't have the necessary hardware, I couldn't test the both the CUDA and ROCM instructions. Any confirmation of it working would be appreciated.

ROCM (AMD)

1. Add -rocm to the end of the Image= line in immich-machine-learning.container.

   - Image=ghcr.io/immich-app/immich-server:${IMMICH_VERSION}
   + Image=ghcr.io/immich-app/immich-server:${IMMICH_VERSION}-rocm

2. Add the following lines in immich-machine-learning.container under [Container].

   GroupAdd=video
   AddDevice=/dev/dri
   AddDevice=/dev/kfd
   

   You may need to add Environment=HSA_OVERRIDE_GFX_VERSION="x.x.x" depending on your GPU.

CUDA (NVIDIA)

1. Add -cuda to the end of the Image= line in immich-machine-learning.container.

   - Image=ghcr.io/immich-app/immich-server:${IMMICH_VERSION}
   + Image=ghcr.io/immich-app/immich-server:${IMMICH_VERSION}-cuda

2. Add the following lines in immich-machine-learning.container under [Container].

   AddDevice=/dev/dri
   # Make sure this matched your GPU ID
   AddDevice=nvidia.com/gpu=0
   Unmask=/dev/dri:/dev/dri
   

---

thanks to tbelways repo for inspiration