Exception: eglInitialize failed (0x3001) #8
Comments
|
Hi @O-O1024, where are you running ComfyUI in this case? Locally or in the cloud? Others have experienced issues running this nodepack on Runpod due to not having OpenGL acceleration available which will cause the same error message above. |
I running it in the cloud with ssh tool. |
|
@O-O1024 It's definitely a driver installation issue (or the lack of) in the host server and/or docker image you're running It's certainly possible to run it on the cloud, for example this Google Colab notebook (slow, no gl accel), or modal.com script file I made, and also this community member that got it running on Inferless (apparently worked? #notsponsored)
EGL is used for headless rendering without the use of a video server (X11/Wayland on Linux). If you want to disable it, set If you have the option to modify or start off a custom docker image in your provider, then you're maybe in luck to get GPU acceleration, otherwise speeds will be as bad as a couple seconds per frame as seen in colab. The most important thing; the host system must have the The general idea is to start with the There'll probably be quirks on each provider, for example, in modal one needs to ask their support team to move the workspace to their older runner which supports the I know talk is cheap, but without knowing what exact provider or hardware you're trying to run, it's hard to pin down to something specific, so I mainly linked resources and TLDR it here :) But ultimately, you're missing any of Let us know if any of this works! 🙂 |
|
I investigated this today; starting off a local docker images with After investigating what magic # (Ubuntu 22.04)
RUN apt update && apt install -y libegl1-mesa libglvnd-dev libglvnd0
RUN mkdir -p /usr/share/glvnd/egl_vendor.d
RUN echo '{"file_format_version":"1.0.0","ICD":{"library_path":"/usr/lib/x86_64-linux-gnu/libEGL_nvidia.so.0"}}' > /usr/share/glvnd/egl_vendor.d/10_nvidia.json
# Could also be important
ENV NVIDIA_DRIVER_CAPABILITIES="all"
ENV NVIDIA_VISIBLE_DEVICES="all"All three packages were required for my RTX 3060 to pop up as a OpenGL renderer, and the ICD file is needed to list the nvidia as a EGL capable device; you can modify or run these commands on a base image in runpod or similar providers, or connect via SSH or remote VNC access to execute the commands directly! Note: The provider must enable let us know if this works for whatever provider you might be running! |
⭐️ Big news, what I said on my previous comment seems to work for runpod!Community member screenshot:
So yea, installing the three packages and the icd file did the trick! We now have a cloud provider with OpenGL acceleration working, without needing to ask their support team, with only minor intervention needed 🙂 |
|
Closing this as it appears to be fixed based on the docker code provided in the mentioned issue above. |
|
|
|
I don't know how to solve it.
The text was updated successfully, but these errors were encountered: