NVIDIA-SMI and Nvidia Runtime Driver don't match

[valentinhendrik]

Hello,

I set up a WSL2 environment to run neural network training and I tried to install Cuda 11.7 for this. But if I run “nvidia-smi” it all looks messed up and I don’t know how to continue. While loading my neural network I also get errors.

libEGL warning: failed to open /dev/dri/renderD128: Permission denied

libEGL warning: failed to open /dev/dri/renderD128: Permission denied

Unable to initialize EGL

Do I get the errors because of the messed up drivers? Am I supposed to set up a whole new WSL2 system? I am pretty new to all of this stuff so I am pretty sure I did a mistake at any point, but I don’t know where and how to fix it.

Here is the nvidia-smi and nvcc --version output:

kiropro@DESKTOP-JC8EGB7:~/PoseCNN-PyTorch$ nvidia-smi
Tue Feb 21 15:20:39 2023
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 525.89.02    Driver Version: 528.49       CUDA Version: 12.0     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  On   | 00000000:01:00.0  On |                  Off |
|  0%   34C    P8    15W / 450W |   1323MiB / 24564MiB |      2%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|    0   N/A  N/A        27      G   /Xwayland                       N/A      |
+-----------------------------------------------------------------------------+
kiropro@DESKTOP-JC8EGB7:~/PoseCNN-PyTorch$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Wed_Jun__8_16:49:14_PDT_2022
Cuda compilation tools, release 11.7, V11.7.99
Build cuda_11.7.r11.7/compiler.31442593_0

When I try to install the graphic driver from Nvidia directly I get following messages which indicate that CUDA 11.7 should be installed:

kiropro@DESKTOP-JC8EGB7:~$ sudo apt-get -y install cuda
Reading package lists... Done
Building dependency tree... Done
Reading state information... Done
cuda is already the newest version (11.7.1-1).
0 upgraded, 0 newly installed, 0 to remove and 0 not upgraded.

EDIT: As I have read 11.7 should be installed, nvidia-smi only shows what my graphic card is capable of, so that should be alright. But still I am confused about the error in the first box and the different NIVIDIA-SMI (525.89.02) driver version and Driver Version (528.49) in the nvidia-smi command

[generix]

You’re just confused by the way nvidia on wsl2 works. There’s no real nvidia gpu, it doesn’t get passed through. Instead, a kind of “cuda proxy” is used to pass the compute instructions from the linux environment on to the windows driver.
Outputs explained:

NVIDIA-SMI 525.89.02

the nvidia-smi version, installed alongside cuda

Driver Version: 528.49

the underlying Windows driver version

CUDA Version: 12.0

The maximum supported cuda version by the driver

Cuda compilation tools, release 11.7

the installed cuda-toolkit version.

Since there’s no real nvidia gpu in wsl2, you must not install a nvidia driver. The state of your current install depends on which repo you used to install the “cuda” metapackage
WSL-Ubuntu repo: doesn’t contain a driver, all fine
Plain Ubuntu repo: not good, “cuda” metapackage includes the driver, “cuda-toolkit” metapackage has to be used instead.

https://docs.nvidia.com/cuda/wsl-user-guide/index.html#getting-started-with-cuda-on-wsl

libEGL warning: failed to open /dev/dri/renderD128: Permission denied

Shouldn’t have anything to do with nvidia, please check its permission:
ls -l /dev/dri

[valentinhendrik]

Okay thank you, now my confusion is gone!

I did install according to the WSL-Ubuntu repo instructions, so that should be fine I guess.

Yes I already checked the permissions, didn’t help :/ But I guess since it has nothing to do with the driver this topic is not suited for the NVIDIA forum.