OpenGL Docker NVidia Win10 - no OpenGL

[j-p-e]

1. Issue or feature description

Win10 docker - OpenGL container doesn't do OpenGL (but can run nvidia-smi and the cuda one can do cuda)

2. Steps to reproduce the issue

#have docker desktop 4.5.1 running on win10 10.0.19044 with WSL2 kernel 5.10.60.1 #on windows host : vcxsrv.exe -multiwindow -clipboard -ac -logfile "%TEMP%\log.txt" -logverbose 5 -screen 0 640x480@ docker run -it -e "DISPLAY=host.docker.internal:0" --privileged --gpus all -e NVIDIA_VISIBLE_DEVICES=all -e NVIDIA_DRIVER_CAPABILITIES=all nvidia/cudagl:11.4.2-runtime-ubuntu20.04

# note same thing happens for opengl:1.2-glvnd-runtime-ubuntu20.04 (nominally built into the one above) # [some flags probably obsolete, I have been trying hundreds of combos from random internet pages] # [yes, I want to send display via a docker IP address - otherwise later on I can't get from gpu container to other containers on # same docker network due to docker host network limitations

# in the docker container created by docker run: nvidia-smi
# OK, can find NVidia-SMI 510.47.03 driver 511.65 Cuda 11.6 on Nvidia GeForce RTX 3080 Ti apt-get update && apt-get install mesa-utils glxinfo -B # Vendor is nvidia, but libGL is not working : libGL: MESA-LOADER: dlopen(/usr/lib/x86_64-linux-gnu/dri/swrast_dri.so) libGL: Can't open configuration file /etc/drirc: No such file or directory. libGL: Can't open configuration file /root/.drirc: No such file or directory. libGL: Can't open configuration file /etc/drirc: No such file or directory. libGL: Can't open configuration file /root/.drirc: No such file or directory. direct rendering: No libGL error: No matching fbConfigs or visuals found libGL error: failed to load driver: swrast OpenGL vendor string: NVIDIA Corporation OpenGL renderer string: NVIDIA GeForce RTX 3080 Ti/PCIe/SSE2 OpenGL version string: 1.4 (4.6.0 NVIDIA 511.65)

# If I run with LIBGL_ALWAYS_INDIRECT in the container I get even less (GPU continues to be visible to nvidia-smi but no # longer visible to glxinfo which defaults to llvmpipe), which sort of makes sense. If I run with -nowgl on the X server outside # the container I also lose visibility of the graphics card from glxinfo

# Expected result : # can use OpenGL from opengl:1.2-glvnd-runtime-ubuntu20.04 or nvidia/cudagl:11.4.2-runtime-ubuntu20.04

# What I actually want to do : # run unity within the container (all on the same actual machine as docker, vcxsrv, and the GPU itself), which I can do already # but only get CPU accel - but have GPU accel work, possibly using unity flags such as -force-opengl

# side note that might help ppl - an apt-get upgrade will kill even this setup so that glxinfo cannot find NVidia at all apt-get update && apt-get -qq upgrade glxinfo -B # Not OK Vendor is Mesa/X.org v12.2.6, Device llvmpipe, Accelerated: no

[j-p-e]

I appreciate this might not quite an nvidia-docker question as I'm using gpus -all - still not found the magic combo that works. Some other things I've tried :

• move the X server to the container in case the X server looking remote is making LIBGL do something unexpected, starting from nvidia/cudagl:11.4.2-runtime. When trying to add in GLX, XOrg always fails with 'cannot create screen' on loading the nvidia driver (with dummy screen, no connected monitor, various validation checks removed, use display device none, have tried a bunch of other nvidia-xconfig options, and I can't find a combo of the libraries that gpus -all hardcodes in and an nvidia library install like xserver-xorg-video-nvidia-510 or ./NVIDIA-Linux-x86_64-510.47.03.run that will use an nvidia glx driver and sit nicely, even with manual relinking of symlinks
• run with ehfd/nvidia-egl-desktop:latest - this works to host.docker.internal:0 screens, does some egl stuff, has 'rendering direct' (but not 'accelerated') and is even clearly using the GPU for some desktop things, but chooses the s/w mode for OpenGL or Unity3D, with or without vglrun
• run with ehfd/docker-nvidia-glx-desktop - this looks like it used to work, but now clashes on docker startup as libnvidia-ml.so.1 and 4 other drivers are already there so won't start, and also looks for something at /proc/devices/nvidia or similar that isn't there any more. If you hack it until it does start, then GLX has a software pipe and is not accelerated
• running with and without --privileged
• tried various things into xorg.conf, am using a PCI Bus ID extracted from nvidia-smi
• https://github.com/trn84/recipe-wizard/blob/master/nvidia-headless.md - this uses TurboVNC to launch a local X server, and used to work 2 years ago, but is now out of date, if you update everything in it you get extension "GLX" missing on display, which is a bug first noticed in 2004 (!) on a gentoo forum
• my NVIDIA-SMI within the container does gives versions NVIDIA-SMI 510.47.03 Driver Version: 511.65 and I've seen that these maybe ought to line up - but the nvidia studio windows release numbers that is giving one of these seem orthogonal with the linux ones I have to install in the container to avoid getting a mesa driver, so I can't see how to do that, I've chosen the closest pair I could find

If anyone from nvidia or anywhere else knows of a set of nvidia driver and X server combo that works with "docker run gpus -all" on a windows host to allow unity3D to accelerate within a linux docker container using a headless display (with or without coming up on the monitor connected to the windows host, it's a simulation so I don't care), please shout

[j-p-e]

Update - I have moved to docker-ce within WSL2, running a docker container via nvidia-docker-container with ocker run --gpus all -it --entrypoint /bin/bash --privileged --gpus all -e NVIDIA_VISIBLE_DEVICES=all -e NVIDIA_DRIVER_CAPABILITIES=all

and

trying to get the nvidia driver rather than mesa in the docker container, here is an nvidia-container-toolkit.log from the WSL2 side

nvidia-container-toolkit.log

and glxinfo from inside the container (noting to get this I have to install mesa-utils) apt-get update && apt-get install mesa-utils && glxinfo -B

direct rendering: Yes Extended renderer info (GLX_MESA_query_renderer): Vendor: Mesa/X.org (0xffffffff) Device: llvmpipe (LLVM 13.0.0, 256 bits) (0xffffffff) Version: 22.0.0 Accelerated: no Video memory: 25601MB Unified memory: no Preferred profile: core (0x1) Max core profile version: 4.5 Max compat profile version: 4.5 Max GLES1 profile version: 1.1 Max GLES[23] profile version: 3.2 OpenGL vendor string: Mesa/X.org OpenGL renderer string: llvmpipe (LLVM 13.0.0, 256 bits) OpenGL core profile version string: 4.5 (Core Profile) Mesa 22.0.0 - kisak-mesa PPA OpenGL core profile shading language version string: 4.50 OpenGL core profile context flags: (none) OpenGL core profile profile mask: core profile

OpenGL version string: 4.5 (Compatibility Profile) Mesa 22.0.0 - kisak-mesa PPA OpenGL shading language version string: 4.50 OpenGL context flags: (none) OpenGL profile mask: compatibility profile

OpenGL ES profile version string: OpenGL ES 3.2 Mesa 22.0.0 - kisak-mesa PPA OpenGL ES profile shading language version string: OpenGL ES GLSL ES 3.20

[j-p-e]

for completeness - glxinfo -B from WSL2 ubuntu

direct rendering: Yes Extended renderer info (GLX_MESA_query_renderer): Vendor: Microsoft Corporation (0xffffffff) Device: D3D12 (NVIDIA GeForce RTX 3080 Ti) (0xffffffff) Version: 21.2.6 Accelerated: yes Video memory: 28457MB Unified memory: no Preferred profile: core (0x1) Max core profile version: 3.3 Max compat profile version: 3.1 Max GLES1 profile version: 1.1 Max GLES[23] profile version: 3.0 OpenGL vendor string: Microsoft Corporation OpenGL renderer string: D3D12 (NVIDIA GeForce RTX 3080 Ti) OpenGL core profile version string: 3.3 (Core Profile) Mesa 21.2.6 OpenGL core profile shading language version string: 3.30 OpenGL core profile context flags: (none) OpenGL core profile profile mask: core profile

OpenGL version string: 3.1 Mesa 21.2.6 OpenGL shading language version string: 1.40 OpenGL context flags: (none)

OpenGL ES profile version string: OpenGL ES 3.0 Mesa 21.2.6 OpenGL ES profile shading language version string: OpenGL ES GLSL ES 3.00

[j-p-e]

also nvidia-smi from inside container NVIDIA-SMI 510.60.02 Driver Version: 512.15 CUDA Version: 11.6

and in WSL2 NVIDIA-SMI 510.60.02 Driver Version: 512.15 CUDA Version: 11.6

[j-p-e]

and symlink setup

root@f6b9d21b568a:/# cd /usr/lib/x86_64-linux-gnu && ls -lA libGL* lrwxrwxrwx 1 root root 14 Nov 19 19:43 libGL.so.1 -> libGL.so.1.7.0 -rw-r--r-- 1 root root 538808 Nov 19 19:43 libGL.so.1.7.0 lrwxrwxrwx 1 root root 32 Mar 23 10:39 libGLESv1_CM_nvidia.so.1 -> libGLESv1_CM_nvidia.so.510.60.02 -rwxr-xr-x 1 root root 67880 Mar 22 21:53 libGLESv1_CM_nvidia.so.510.60.02 lrwxrwxrwx 1 root root 18 Nov 19 19:43 libGLESv2.so.2 -> libGLESv2.so.2.1.0 -rw-r--r-- 1 root root 63664 Nov 19 19:43 libGLESv2.so.2.1.0 lrwxrwxrwx 1 root root 29 Mar 23 10:39 libGLESv2_nvidia.so.2 -> libGLESv2_nvidia.so.510.60.02 -rwxr-xr-x 1 root root 117032 Mar 22 21:53 libGLESv2_nvidia.so.510.60.02 lrwxrwxrwx 1 root root 15 Nov 19 19:43 libGLX.so.0 -> libGLX.so.0.0.0 -rw-r--r-- 1 root root 133568 Nov 19 19:43 libGLX.so.0.0.0 lrwxrwxrwx 1 root root 16 Mar 10 20:56 libGLX_indirect.so.0 -> libGLX_mesa.so.0 lrwxrwxrwx 1 root root 20 Mar 10 20:56 libGLX_mesa.so.0 -> libGLX_mesa.so.0.0.0 -rw-r--r-- 1 root root 467608 Mar 10 20:56 libGLX_mesa.so.0.0.0 lrwxrwxrwx 1 root root 26 Mar 23 10:39 libGLX_nvidia.so.0 -> libGLX_nvidia.so.510.60.02 -rwxr-xr-x 1 root root 1293712 Mar 22 21:53 libGLX_nvidia.so.510.60.02 lrwxrwxrwx 1 root root 22 Nov 19 19:43 libGLdispatch.so.0 -> libGLdispatch.so.0.0.0 -rw-r--r-- 1 root root 710944 Nov 19 19:43 libGLdispatch.so.0.0.0

[j-p-e]

and am now launching vcsrv from WSL2 with

/mnt/c/program\ files/vcxsrv/vcxsrv.exe :1974 -listen tcp -retro -lesspointer -nowgl +iglx -multiwindow -clipboard -screen 0 '640x480' -ac +extension RANDR +extension RENDER +extension GLX +extension DOUBLE-BUFFER +extension DAMAGE +extension COMPOSITE -extension X-Resource -extension XTEST

[j-p-e]

and on WSL dpkg -l | grep libnvidia ii libnvidia-common-510 510.54-0ubuntu0.20.04.1 all Shared files used by the NVIDIA libraries ii libnvidia-compute-510:amd64 510.47.03-0ubuntu0.20.04.1 amd64 NVIDIA libcompute package ii libnvidia-container-tools 1.9.0-1 amd64 NVIDIA container runtime library (command-line tools) ii libnvidia-container1:amd64 1.9.0-1 amd64 NVIDIA container runtime library ii libnvidia-gl-510:amd64 510.54-0ubuntu0.20.04.1 amd64 NVIDIA OpenGL/GLX/EGL/GLES GLVND libraries and Vulkan ICD

[lovesuiyue]

Hello, have you installed The entire emulator yet?