[SOLVED] Blender Ubuntu not detecting new GPU after switch

Support Technical Support
#1

Hello again,

So I recently bought 2 Nvidia p106-100 cards (the headless ones meant for mining) to use with Blender. The first one I put in my main rig next to my GTX-1070 running Win10, and it worked straight out of the box. All I had to do was tick the checkbox and bam I had a GTX-1060 for about $70

However, I put the second one in my Headless render server this morning and ran into an issue. I used to have a GTX-970 in it, but when I swapped it out with the p106, and run

import bpy
prefs = bpy.context.user_preferences.addons['cycles'].preferences
devices = prefs.devices
print(prefs.compute_device_type)    # Prints out if CUDA is enabled

for d in devices:
        print(d.name)    # Prints out each device

I get the following output

CUDA                                 # CUDA is enabled
AMD FX(tm)-6350 Six-Core Processor   # correct system processor 
GeForce GTX 970 (Display)            # Previous graphics card that isn't even part of the system anymore

I know the system is detecting the p106 because it’s listed when I run lspci -v


01:00.0 3D controller: NVIDIA Corporation GP106 [P106-100] (rev a1)
        Subsystem: Gigabyte Technology Co., Ltd GP106 [P106-100]
        Flags: bus master, fast devsel, latency 0, IRQ 33, NUMA node 0
        Memory at fd000000 (32-bit, non-prefetchable) [size=16M]
        Memory at c0000000 (64-bit, prefetchable) [size=256M]
        Memory at d0000000 (64-bit, prefetchable) [size=32M]
        Capabilities: <access denied>
        Kernel driver in use: nvidia
        Kernel modules: nvidiafb, nouveau, nvidia_drm, nvidia

It’s still showing the GTX-970 as my graphics card, but I am literally staring at the 970 sitting on the table next to me. I’ve tried rebooting, and reinstalling the Nvidia and CUDA drivers, but same issue. Any Ideas?

#2

Perhaps it’s the device listed as ‘CUDA’ ?

#3

Sorry, perhaps I should have explained that script better. The first line isn’t a device, it’s just checking to see if CUDA is enabled, which it is. The FX-6350 is the CPU, and the GTX-970 is the graphics card that isn’t even part of the system anymore.

#4

It seems that Blender doesn’t erase old devices from wherever it stores them. I used to have the same GTX-970 that is in the render rig in my main workstation rig, so when I run the same code on my Win10 machine it says that I have:

CUDA
GeForce GTX 970 (Display)
GeForce GTX 1070 (Display)
Intel Core i7-6700K CPU @ 4.00GHz
P106-100

So on Win10 it lists the p106 card with no problem, but also displays the old 970 card that isn’t in the system anymore. Running with “–factory-startup” seems to remove the old card, so I will play around with it later when I have time to see if I can get it working.

#5

Ok, figured it out. Thankfully on my remote machine I didn’t install any addons or change the user preferences, so I just ran the following

blender -b --python-console              # Start python console
>> import bpy
>> bpy.ops.wm.read_factory_settings()    # Load factory settings
>> bpy.ops.wm.save_userpref()            # Save factory settings

I was then able to open the python console again and simply set the CUDA device and graphics card using the method from this thread like I did before: Thread

After doing this the script I used displayed the following (correct) results:

CUDA
AMD FX(tm)-6350 Six-Core Processor
P106-100

And GPU rendering worked perfectly.