# Headless render: can't find compute_device in 2.78b

On Ubuntu 16.04 (used as a headless renderer with dual GTX 970 STRIX GPUs) after upgrading to blender 2.78b I'm now seeing:

$blender -b Lightsaber.blend -o //Lightsaber-tst-1_ -E CYCLES -F PNG -x 1 -P cuda0.py -f 1 AL lib: (WW) alc_initconfig: Failed to initialize backend "pulse" ALSA lib pcm_dmix.c:1029:(snd_pcm_dmix_open) unable to open slave AL lib: (EE) ALCplaybackAlsa_open: Could not open playback device 'default': No such file or directory read blend: /home/shady/usermaatre/School/Lightsaber.blend Traceback (most recent call last): File "/home/shady/usermaatre/School/cuda0.py", line 4, in <module> bpy.context.user_preferences.system.compute_device_type = 'CUDA'  The cuda0.py script is simple (it comes from BlenderArtists): $ cat cuda0.py
import bpy, _cycles

bpy.context.scene.cycles.device = 'GPU'
bpy.context.user_preferences.system.compute_device_type = 'CUDA'

# this is different in each cuda[x].py file, CUDA_0, CUDA_1, CUDA_2, CUDA_3
bpy.context.user_preferences.system.compute_device = 'CUDA_0'


I had been running NVIDIA-Linux-x86_64-367.44.run and cuda_7.5.18_linux.run so we tried upgrading to the latest NVIDIA-Linux-x86_64-375.39.run and cuda_8.0.61_375.26_linux.run but that did not help.

I figured maybe the CUDA card has been renamed, so I tried running the script at How do I get Cycles to use GPU on a linux server? and I get a similar result:

\$ blender -b -P compute_device_type.py
AL lib: (WW) alc_initconfig: Failed to initialize backend "pulse"
ALSA lib pcm_dmix.c:1029:(snd_pcm_dmix_open) unable to open slave
AL lib: (EE) ALCplaybackAlsa_open: Could not open playback device 'default': No such file or directory
Traceback (most recent call last):
File "/home/steven/usermaatre/School/compute_device_type.py", line 7, in <module>
devt = sysp.compute_device_type
AttributeError: 'UserPreferencesSystem' object has no attribute 'compute_device_type'

Blender quit


I'm thinking that something has changed in the naming of the card? Is there a simple way to determine the card attributes? And should the 8.0 drivers work on a headless system or should I go back to 7.5? I'd really rather not have to drop back to 2.78a which had worked...

From the 2.78b, it's necessary to recover them via the preferences of the Cycles addon. You cant get it with

bpy.context.user_preferences.addon['cycles'].preferences.compute_device_type


And for the devices:

for card in bpy.context.user_preferences.addons['cycles'].preferences.devices:
print(card.name)

• Also of note - compute_device_type is an enum - so if you try compute_device_type = 'XXX' you will get a TypeError message that lists the valid enum values. – sambler Feb 21 '17 at 4:48
• @sambler @pistiwique Thanks, I get enum "XXX" not found in ('NONE', 'CUDA') so setting prefs.compute_device_type = 'CUDA' works and both cards are used. I would also like to be able to address GPU0 & GPU1 separately, but prefs.compute_device = 'CUDA_0' fails. – Shady Puck Feb 21 '17 at 23:35
• I haven't tested this but it appears that the devices list shown in the above loop, is what you use to enable/disable devices. Set compute_device_type = 'CUDA' and then use addons['cycles'].preferences.devices[0].use = False to disable the first device and devices[1] for the second. Each device has id, name, type and use properties. – sambler Feb 22 '17 at 0:45
• @sambler thanks - prefs.devices[0].use = False worked! The above loop printed out <bpy_struct, CyclesDeviceSettings("GeForce GTX 970")> twice but I was not able to determine the deeper structure. How did you determine to set .use = False (where the old way was .compute_device = 'CUDA_0')? I wonder what other options might exist? – Shady Puck Feb 22 '17 at 20:46
• I used the source @ShadyPuck ;) - You can find CyclesDeviceSettings defined in properties.py inside the cycles addon. I expect this is a simpler approach to controlling the many variations of different devices. It has just been added so expect more changes to come. – sambler Feb 23 '17 at 3:12

It seems this method doesnt work with 2.79? I get the error:

AttributeError: 'UserPreferences' object has no attribute 'addon'

Edit: Got it working! a stupid missing 's'

Here is the working script i'm using to force the worker station to use all cuda devices:

import bpy