Blender Google Colab | 易学教程

问题

Up until few days ago I was able to run Blender(or sheep it) headless on Google Colab with an GPU(yes I checked trice that I selected a GPU Instance). Now running keras/tensorflow does work fine but blender can't detect the GPU at all and Sheepit tells me that

"cuInit Failed ret: 100"

nvidia-smi results with:

Sat Aug 24 19:48:06 2019       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 430.40       Driver Version: 418.67       CUDA Version: 10.1     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  Tesla K80           Off  | 00000000:00:04.0 Off |                    0 |
| N/A   33C    P8    25W / 149W |      0MiB / 11441MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

The following script did work fine for few months and detected the GPU properly and setup blender to use the available GPU. https://gist.github.com/donmahallem/a05100077ec1327268f28f0b2bd8da60

I tried a lot but my nvidia/unix foo is pretty limited. I tried to find out if the CUDA Version changed and tried to downgrade the cuda version to 9.2 but couldn't get it to work again.

回答1:

I'm also facing the same problem. I think colab is not allowing GPU to use for any commands that include "sudo" in it.

Solution I found was to install blender and then run the command to render without "sudo".

I installed blender from PPA ppa:thomas-schiex/blender

commands to execute:

!sudo add-apt-repository ppa:thomas-schiex/blender
!sudo apt-get install blender
!blender -b -P script.py filename.blend -o fileoutput -F PNG -f 1

script.py contains code to select gpu, I generally upload to colab. Save following code as script.py

import bpy, _cycles

bpy.context.scene.cycles.device = 'GPU'

avail_devices = _cycles.available_devices('CUDA')
print(avail_devices)

prop = bpy.context.preferences.addons['cycles'].preferences

prop.get_devices(prop.compute_device_type)
prop.compute_device_type = 'CUDA'

for device in prop.devices:
    if device.type == 'CUDA':
        print('device: ', device)
        device.use = True

This worked for me.

来源：https://stackoverflow.com/questions/57641141/blender-google-colab

标签

blender

google-colaboratory