Debian Linux Blender can't see CUDA devices, says "CUEW initialization failed" #84802

New Issue

Brenden · 2021-01-17T17:51:55+01:00

Brenden commented

2021-01-17 17:51:55 +01:00

System Information
Operating system: Debian GNU/Linux 10 (buster)
Graphics card: Nvidia GeForce GTX 1070

Blender Version
Broken: 2.83.7, 2.90.1, 2.91.0
Worked: unsure
Short description of error

Blender is unable to use my Nvidia GPU for Cycles rendering. I have Nvidia drivers working, and I noticed that Eevee renders use the Nvidia GPU properly, as I can see 99% GPU utilization during Eevee renders using the nvidia-smi command.

Under Edit -> Preferences -> System -> CUDA, an error message says "No compatible GPUs found for path tracing Cycles will render on the CPU".

If I run blender with Cycles debugging on (blender --debug-cycles), I see an error related to opening a CUDA library:
I0117 08:45:47.952724 1230 device_cuda.cpp:56] CUEW initialization failed: Error opening the library

Exact steps for others to reproduce the error

Download the newest Blender (2.91.0).
Extract the tar.bz2 file to a directory somewhere.
From within the new directory, run blender --debug-cycles & to start blender in the background with Cycles debugging on
Navigate to Edit -> Preferences -> System and then choose CUDA under Cycles Render Devices
You should see a message that no compatible GPUs are available.
The debug logs in the terminal should show the CUEW initialization failed: Error opening the library error.

Output of nvidia-smi command
Below is the output of nvidia-smi, which shows driver versions, etc. in case that helps. Just as an example of Eevee using the GPU successfully, this output of nvidia-smi shows 99% utilization during a computationally intensive Eevee render:

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 418.152.00   Driver Version: 418.152.00   CUDA Version: N/A      |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 1070    On   | 00000000:02:00.0  On |                  N/A |
|  0%   55C    P0   101W / 200W |   2578MiB /  8119MiB |     99%      Default |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
| -- |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0      1099      G   blender                                     1917MiB |
|    0      1387      G   /usr/lib/xorg/Xorg                            26MiB |
|    0      1567      G   /usr/bin/gnome-shell                          25MiB |
|    0      1849      G   /usr/lib/xorg/Xorg                           275MiB |
|    0      1926      G   /usr/bin/gnome-shell                         276MiB |
|    0      4967      G   .../share/spotify/debug.log --shared-files    54MiB |
+-----------------------------------------------------------------------------+

**System Information** Operating system: Debian GNU/Linux 10 (buster) Graphics card: Nvidia GeForce GTX 1070 **Blender Version** Broken: 2.83.7, 2.90.1, 2.91.0 Worked: unsure **Short description of error** Blender is unable to use my Nvidia GPU for Cycles rendering. I have Nvidia drivers working, and I noticed that Eevee renders use the Nvidia GPU properly, as I can see 99% GPU utilization during Eevee renders using the `nvidia-smi` command. Under `Edit -> Preferences -> System -> CUDA`, an error message says "No compatible GPUs found for path tracing Cycles will render on the CPU". If I run blender with Cycles debugging on (`blender --debug-cycles`), I see an error related to opening a CUDA library: `I0117 08:45:47.952724 1230 device_cuda.cpp:56] CUEW initialization failed: Error opening the library` **Exact steps for others to reproduce the error** 1. Download the newest Blender (2.91.0). 2. Extract the tar.bz2 file to a directory somewhere. 3. From within the new directory, run `blender --debug-cycles &` to start blender in the background with Cycles debugging on 4. Navigate to `Edit -> Preferences -> System` and then choose `CUDA` under `Cycles Render Devices` 5. You should see a message that no compatible GPUs are available. 6. The debug logs in the terminal should show the `CUEW initialization failed: Error opening the library` error. **Output of nvidia-smi command** Below is the output of `nvidia-smi`, which shows driver versions, etc. in case that helps. Just as an example of Eevee using the GPU successfully, this output of `nvidia-smi` shows 99% utilization during a computationally intensive Eevee render: ``` +-----------------------------------------------------------------------------+ | NVIDIA-SMI 418.152.00 Driver Version: 418.152.00 CUDA Version: N/A | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | |===============================+======================+======================| | 0 GeForce GTX 1070 On | 00000000:02:00.0 On | N/A | | 0% 55C P0 101W / 200W | 2578MiB / 8119MiB | 99% Default | +-------------------------------+----------------------+----------------------+ +-----------------------------------------------------------------------------+ | Processes: GPU Memory | | -- | | GPU PID Type Process name Usage | |=============================================================================| | 0 1099 G blender 1917MiB | | 0 1387 G /usr/lib/xorg/Xorg 26MiB | | 0 1567 G /usr/bin/gnome-shell 25MiB | | 0 1849 G /usr/lib/xorg/Xorg 275MiB | | 0 1926 G /usr/bin/gnome-shell 276MiB | | 0 4967 G .../share/spotify/debug.log --shared-files 54MiB | +-----------------------------------------------------------------------------+ ```

Brenden commented

2021-01-17 17:51:55 +01:00

Added subscriber: @brendenahyde

Alaska commented

2021-01-17 22:07:13 +01:00

Added subscriber: @Alaska

Alaska commented

2021-01-17 22:07:13 +01:00

What version of the Linux Kernel are you running? You can find out which version you're using with uname -r in the terminal/shell/etc.

I ask this as with Kernel version 5.9 and higher, CUDA no longer works on Nvidia GPUs unless you are running driver version 455.45 or higher.

What version of the Linux Kernel are you running? You can find out which version you're using with `uname -r` in the terminal/shell/etc. I ask this as with Kernel version 5.9 and higher, CUDA no longer works on Nvidia GPUs unless you are running driver version 455.45 or higher.

Alaska commented

2021-01-17 22:08:12 +01:00

Another thing you need to do is ensure you have CUDA libraries installed. This can be done by running the command sudo apt install libcuda1

Another thing you need to do is ensure you have CUDA libraries installed. This can be done by running the command `sudo apt install libcuda1`

Robert Guetzkow commented

2021-01-18 11:30:31 +01:00

Added subscriber: @rjg

Robert Guetzkow commented

2021-01-18 11:30:31 +01:00

Changed status from 'Needs Triage' to: 'Needs User Info'

Robert Guetzkow commented

2021-01-18 11:30:31 +01:00

As @Alaska said, please install the latest Nvidia driver for your GPU.

Brenden commented

2021-01-19 03:43:10 +01:00

@Alaska @rjg thank you both. The issue was the Nvidia driver version as you said. Debian only has up to version 418 available in the normal repos, so I had to install the driver directly from Nvidia's website. This was a risk since it's not vetted by the Debian maintainers yet, but it worked, and my issue is resolved.

For the record, my kernel version is 4.19.0-13-amd64.

Just in case this helps someone else in the future, here is the Nvidia driver version I have now (upgraded from 418.152.00):

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 460.32.03    Driver Version: 460.32.03    CUDA Version: 11.2     |
|-------------------------------+----------------------+----------------------+

I appreciate your help!

@Alaska @rjg thank you both. The issue was the Nvidia driver version as you said. Debian only has up to version 418 available in the normal repos, so I had to install the driver directly from Nvidia's website. This was a risk since it's not vetted by the Debian maintainers yet, but it worked, and my issue is resolved. For the record, my kernel version is `4.19.0-13-amd64`. Just in case this helps someone else in the future, here is the Nvidia driver version I have now (upgraded from `418.152.00`): ``` +-----------------------------------------------------------------------------+ | NVIDIA-SMI 460.32.03 Driver Version: 460.32.03 CUDA Version: 11.2 | |-------------------------------+----------------------+----------------------+ ``` I appreciate your help!

Alaska commented

2021-01-19 09:49:06 +01:00

Changed status from 'Needs User Info' to: 'Resolved'

Alaska closed this issue

2021-01-19 09:49:06 +01:00

Alaska self-assigned this 2021-01-19 09:49:06 +01:00

Alaska commented

2021-01-19 12:02:23 +01:00

Note for @brendenahyde . You can install more recent drivers officially verified by the Debian team using the instructions found here: https://backports.debian.org/Instructions/

I will write up my own instructions here:

Open the terminal and run the command sudo nano /etc/apt/sources.list
A text editor should open up. Add this line somewhere: deb http://deb.debian.org/debian buster-backports main contrib non-free and save the file with Ctrl + X
Run the command sudo apt update to update your sources list.
Run the command sudo apt -t buster-backports install nvidia-driver libcuda1 libnvoptix1 nvidia-settings to install the updated drivers and other useful libraries.

There's a few things to note:
I have no idea how this will work with you installing the drivers straight from Nvidia's website.
In most cases this is NOT the most recent Nvidia driver, but it is usually within a month or so of the most recent.

Note for @brendenahyde . You can install more recent drivers officially verified by the Debian team using the instructions found here: https://backports.debian.org/Instructions/ I will write up my own instructions here: 1. Open the terminal and run the command `sudo nano /etc/apt/sources.list` 2. A text editor should open up. Add this line somewhere: `deb http://deb.debian.org/debian buster-backports main contrib non-free` and save the file with `Ctrl + X` 3. Run the command `sudo apt update` to update your sources list. 4. Run the command `sudo apt -t buster-backports install nvidia-driver libcuda1 libnvoptix1 nvidia-settings` to install the updated drivers and other useful libraries. **There's a few things to note:** I have no idea how this will work with you installing the drivers straight from Nvidia's website. In most cases this is **NOT** the most recent Nvidia driver, but it is usually within a month or so of the most recent.

Woland commented

2021-10-17 07:07:44 +02:00

Added subscriber: @woland.tel

Woland commented

2021-10-17 07:07:44 +02:00

Hope this should be helpful: https://devtalk.blender.org/t/nvidia-prime-is-here-but-cuda-not-found/15226/27?u=woland

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Debian Linux Blender can't see CUDA devices, says "CUEW initialization failed" #84802