Add OpenCL support to the container #3833

Activity

1. WesleyFister

   added

   enhancement

   status/needs triage

   on Dec 26, 2025
2. itzg commented on Dec 26, 2025

   

   itzg

   on Dec 26, 2025

   Owner

   More actions

   My only concern would be the increase in image size. Will just need to try it and measure with docker history before and after.

   FYI, this was added as a means for custom images to be built

   https://docker-minecraft-server.readthedocs.io/en/latest/misc/building/#installing-extra-packages
3. itzg

   removed

   status/needs triage

   on Dec 26, 2025
4. itzg

   added this to Docker Minecrafton Dec 26, 2025
5. mattcam98 commented on Feb 23, 2026

   

   mattcam98

   on Feb 23, 2026

   More actions

   I've successfully implemented OpenCL support for the C2ME mod using a custom image. Total image size increase is only ~2MB (851MB → 853MB) because it uses the generic ICD loader rather than bundling heavy driver packages.

   For anyone looking to implement this, here is the working configuration:

   1. Dockerfile

   The key is installing the generic OpenCL loader and manually pointing it to the NVIDIA library that the Docker runtime injects.

   FROM itzg/minecraft-server:java25
   
   # Install the generic OpenCL loader (very lightweight)
   RUN apt-get update && \
       apt-get install -y --no-install-recommends ocl-icd-libopencl1 && \
       rm -rf /var/lib/apt/lists/*
   
   # Create the vendor hook so the loader finds the NVIDIA driver injected by the host
   RUN mkdir -p /etc/OpenCL/vendors && \
       echo "libnvidia-opencl.so.1" > /etc/OpenCL/vendors/nvidia.icd
   

   2. Docker Compose requirements

   To "link" the hardware to the container, the following environment variables and reservations are required:

   services:
     mc:
       image: mc-opencl:latest
       runtime: nvidia
       environment:
         NVIDIA_VISIBLE_DEVICES: all
         NVIDIA_DRIVER_CAPABILITIES: compute,utility
       deploy:
         resources:
           reservations:
             devices:
               - driver: nvidia
                 count: all
                 capabilities: [gpu]
   

   This setup resolves the OpenCL error [-1001] (CL_PLATFORM_NOT_FOUND_KHR) by ensuring the Java/LWJGL layer has a valid ICD path to the host GPU.