OpenCL Issues

[jrowinski-irreventlabs]

Hi there,

Thanks for setting up this repo, been trying to use it and so far it has been great. I've been trying to use a GPU in the container to expose to hou, but for some reason I cannot get it to work. I keep getting errors with OpenCL.

OpenCL Exception: No OpenCL platform found. (-32)
No OpenCL platform has the specified device type (HOUDINI_OCL_DEVICETYPE): GPU.

Running clinfo within the container it returns there are no platforms to use.

I am wondering if anyone has a successful approach to exposing and using a GPU ? Some of the variables I've been playing with are:

      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=all
      - HOUDINI_OCL_DEVICETYPE=GPU
      - HOUDINI_OCL_VENDOR=NVIDIA Corporation
      - HOUDINI_USE_HFS_OCL=0

And yes, I've installed the appropriate drivers, cuda, and container toolkit. I will continue troubleshooting and if I get anything successful I will update this ticket.

Thanks in advance

[jrowinski-irreventlabs]

And just like that, I found the issue to be with NCT -> NVIDIA/nvidia-container-toolkit#682

[aaronsmithtv]

Hey @jrowinski-irreventlabs, thanks for submitting this, and I'm glad you're enjoying Houdini-Docker! Is the bug with NCT something that can be fixed on my end, or am I good to close this?

[jrowinski3d]

Hey @aaronsmithtv , back from the holidays now. I think you would need to do the following like listed in the nvidia issue. I suspect that once they fix the issue with the nvidia toolkit you wont need to do that anymore.

I got mine worked by mounting my like nvidia-icd to the docker container through a volume and it resolved it for me while I test things out

$ cat opencl.Dockerfile 
FROM ubuntu:22.04

ARG DEBIAN_FRONTEND=noninteractive
RUN apt-get update && apt-get install -y clinfo && apt-get clean

RUN install -d /etc/OpenCL/vendors/ && \
    echo 'libnvidia-opencl.so.1' >> /etc/OpenCL/vendors/nvidia.icd

[jrowinski3d]

To add, setting this env var helps alleviate the issue to gaining access to a GPU. Just came across this issue again running the container in a K8s cluster

- name: LD_LIBRARY_PATH
  value: "/usr/local/nvidia/lib64:${LD_LIBRARY_PATH}"