Platform::WindowlessEglApplication::tryCreateContext(): unable to find CUDA device 0 among 2 EGL devices in total

[28dvd28]

Hello, I'm trying to use habitat-sim for my project, but when i'm running it i'm getting the error: Platform::WindowlessEglApplication::tryCreateContext(): unable to find CUDA device 0 among 2 EGL devices in total I've looked in some other already opened issue, but I can't find a solution, maybe also because i'm working on a WSL and not directly on linux machine.

The fact that i am using the WSL can it be the reason of why I'm getting the error? If yes can you help me solving this?

[aclegg3]

Hey @28dvd28

I assume this can work, but it sounds like you'll need to do some work to expose and setup the GPU correctly.

I saw this: https://canonical-ubuntu-wsl.readthedocs-hosted.com/en/latest/tutorials/gpu-cuda/

Also take a look at our troubleshooting guide for graphics issues.

Once you've dug in a bit and can provide some more info maybe myself or @0mdc can help with any questions.

[28dvd28]

Hello,

many thanks for the resources that you gave me.

I've looked at both the link that you have sent to me. From the first one I have a question. I need to download the nvidia driver for windows. But i've already done it for gaming, so i downloaded the game ready driver (I think, it has been i while). Do I need to download the Studio Driver?

I've looked to the second link and i've followed the instruction, but i don't know if i have the right output, so:

• running eglinfo I get Above is a lot of more output, similar to the last lines and then I get Device platofrom without nothing, is it a problem? -apt list --installed | grep libglvnd get me is it ok? -I had to create the file 10_nvidia.json in /usr/share/glvnd/egl_vendor.d/, the path that i have to put in "library_path" must be like "../../../../lib/x86_64-linux-gnu/libEGL.so.1" or like "lib/x86_64-linux-gnu/libEGL.so.1"?

In any case i am still having the issue also testing it in the habitat-lab reposotory with the command python examples/example.py. Hope you will be able to help me.

Kind regards

[wsychiu]

Hello @28dvd28 - did you managed to find a fix for this? I am literally having the same issue as you are.

I can run the provided example examples/viewer.py with the skokloster-castle.glb example, but cannot get it to work with Matterport3D dataset (after following the Matterport3D instructions).

nvidia-smi and nvcc --version are all returning up-to-date drivers and toolkit. I also get the same eglinfo: eglInitialize failed after a wall of text.

Hoping someone has a fix as I can't seem to find many solved problems online...

[28dvd28]

Hello @wsychiu, I wasn't able to solve this issue, I think that maybe for some strange reason it is needed a full Linux OS on your machine and with just the WSL, for some strange reason, it is not working.

[wsychiu]

Thanks @28dvd28 - did you get it working on a full Linux install?

[28dvd28]

@wsychiu - Since it was a school project, I thought that it was too much effort change the OS of my PC, so i did not go on trying to solve it, but the PhD student, that was following me on this project, used a Linux OS and on his PC everything worked fine.

[Sebo-the-tramp]

I might be late to the party, but I struggled for about a day with this problem and finally figured out a solution. In my case, I am running the code on a cluster with NO sudo access, and the drivers have already been installed. After going around looking at issues in particular issue 1671 and issue 288 I figured out that the habitat simulator needs a particular file to be placed in a particular place:

/usr/share/glvnd/egl_vendor.d/10_nvidia.json

turns out I didn't have it...

Therefore I downloaded it from https://github.com/openai/mujoco-py/blob/master/vendor/10_nvidia.json as suggested from issue 288 to a place that I will not forget and then since I have no sudo access, the only option was to export the os variable as follows:

export __EGL_VENDOR_LIBRARY_FILENAMES=/the/path/i/will/not/forget/10_nvidia.json

After that, everything works for me.

[fangbq]

I also came across this problem on a remote linux server with 4090 gpu, but without a display. Tried @Sebo-the-tramp 's solution, does not work for me.

The only way out for me is adding a line of code to use cpu for rendering rather than gpu.

[1nibeloflight2]

I might be late to the party, but I struggled for about a day with this problem and finally figured out a solution. In my case, I am running the code on a cluster with NO sudo access, and the drivers have already been installed. After going around looking at issues in particular issue 1671 and issue 288 I figured out that the habitat simulator needs a particular file to be placed in a particular place:

/usr/share/glvnd/egl_vendor.d/10_nvidia.json

turns out I didn't have it...

Therefore I downloaded it from https://github.com/openai/mujoco-py/blob/master/vendor/10_nvidia.json as suggested from issue 288 to a place that I will not forget and then since I have no sudo access, the only option was to export the os variable as follows:

export __EGL_VENDOR_LIBRARY_FILENAMES=/the/path/i/will/not/forget/10_nvidia.json

After that, everything works for me.

I have /usr/share/glvnd/egl vendor.d/10 nvidia.json, but also be error ,then I download 10_nvidia.json to another dir, use

export __EGL_VENDOR_LIBRARY_FILENAMES=/the/path/i/will/not/forget/10_nvidia.json to my dir. and work !!!Perhaps it's because my environment doesn't have the permission to access the file "usr/share/glvnd/egl vendor.d/10 nvidia.json"