google-deepmind
Problems with egl rendering
I am working on a mujoco simulation inside a docker container (using nvidia-docker for GPU access). My simulation use egl for hardware rendering and on my local machine works perfectly. I would like to run my simulation on a server with 3 NVIDIA A40. When I run this script in the same docker image but on the server
import os
os.environ["MUJOCO_GL"]="egl"
os.environ["MUJOCO_EGL_DEVICE_ID"] = "1"
from dm_control import mujoco
I get this error:
ImportError: Cannot initialize a headless EGL display.
Server: Nvidia Driver Version: 525.85.12 CUDA Version: 12.0
Local Machine: Nvidia Driver Version: 525.116.04 CUDA Version: 12.0
If I use osmesa (os.environ["MUJOCO_GL"]="egl") everything works correctly. But I would like to use egl.
I tried to install libnvidia-gl (apt install libnvidia-gl-525) as in #214 and #365. but I get the same error.
I should also add that I do not use NVIDIA Multi-Instance GPU.
Am I doing something wrong here?
Thank you in advance for the support!
This almost definitely means that your Docker container doesn't have permission to access the GPU hardware specified.
Have you tried setting MUJOCO_EGL_DEVICE_ID to 0 ?
I suspected that. It works with MUJOCO_EGL_DEVICE_ID=0.
I would also add that nvidia-smi within the container shows all the gpu's.
And I'm still not using the GPU! What am I missing?
No use Docker and one gpu has this error. libEGL warning: failed to open /dev/dri/renderD128: Permission denied
libEGL warning: failed to open /dev/dri/card0: Permission denied
how to reslove it?
