eGPU not initializing

[jacobgkau]

Tested on a darp6 (customer experiencing the same issue on a galp4.) When we plug in an NVIDIA GPU in the Akitio Node, we do see it listed in lspci:

system76@pop-os:~$ lspci | grep VGA
00:02.0 VGA compatible controller: Intel Corporation Device 9b41 (rev 02)
06:00.0 VGA compatible controller: NVIDIA Corporation GP108 [GeForce GT 1030] (rev a1)

However, nvidia-smi does not see the GPU, and we are not able to load the NVIDIA driver:

system76@pop-os:~$ nvidia-smi
NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.
system76@pop-os:~$ sudo modprobe nvidia
modprobe: ERROR: could not insert 'nvidia': No such device

dmesg shows this output continuously repeated while the GPU is connected:

[  251.993332] nvidia-nvlink: Nvlink Core is being initialized, major device number 237
[  251.993840] NVRM: This PCI I/O region assigned to your NVIDIA device is invalid:
               NVRM: BAR0 is 0M @ 0x0 (PCI:0000:06:00.0)
[  251.993841] NVRM: The system BIOS may have misconfigured your GPU.
[  251.993845] nvidia: probe of 0000:06:00.0 failed with error -1
[  251.993863] NVRM: The NVIDIA probe routine failed for 1 device(s).
[  251.993864] NVRM: None of the NVIDIA devices were initialized.
[  251.993998] nvidia-nvlink: Unregistered the Nvlink Core, major device number 237

[vthg2themax]

vfio now appears to be broken on 20.04 Pop_OS. :-( @tkolo Just a warning friend.

[theironrobin]

I also have a Razer Core and AMD RX 580, i am not able to get this working. on Pop_OS 20.04... darp6... lspci -k | grep -EA3 'VGA|3D|Display' shows both i915 (intel graphics) and amdgpu kernel drivers in use. but actual performance is not good. lots of screen tearing, very slow, and ~13 fps on gaming benchmarks.

EDIT: It seems that by using this egpu switcher I can get it working correctly, BUT, auto-detect feature is not working. So if I want to use egpu I need to do sudo systemctl disable egpu <- disables the auto switch service, just need to do this once sudo egpu-switcher switch egpu <- manually set X11 setting to use egpu. reboot <- with egpu connected Then, to switch back to internal: sudo egpu-switcher switch internal reboot

NOTE: this only seems to work if auto-login is set up which is a shame, privacy/security wise... also my hard drive is not encrypted, so I am not sure how that would affect things.

[jacobgkau]

NOTE: this only seems to work if auto-login is set up which is a shame, privacy/security wise... also my hard drive is not encrypted, so I am not sure how that would affect things.

I've tested with an RX 580 and a galp4. While I can confirm the auto-switching service doesn't work (seems to always switch to internal on boot), and that an Xorg restart is required after using the egpu-switcher switch command since the Xorg configuration file is being changed out, I am able to use the RX 580 as the primary GPU with full-disk encryption and without enabling auto-login. What issue are you seeing when you disable auto-login?

[kocoman2]

can anyone please tell me what code is the "The fix was to increase the amount of hotplug memory reserved for the Thunderbolt bridge" in? thank you

[crawfxrd]

The coreboot configs PCIEXP_HOTPLUG_MEM and PCIEXP_HOTPLUG_PREFETCH_MEM.

[antonkulaga]

Guys, eGPU does not work with my lemur pro (lemp11 if I am not mistaken) with Razer Core and Nvidia RTX3060. I tried pretty much everything I found on the internet. Has anybody managed to make it work? I bought lemur specifically to bundle it with egpu when I need to do deep-learning and safe charge otherwise, but it does not work :((((

[jacobgkau]

lemp12 isn't initializing an RTX GPU in an Akitio Node Titan. dmesg output: lemp12-egpu.txt (Just trying to plug it in after boot in order to use it for CUDA. nvidia-smi is never showing it.)

A lemp12 customer with a Razer Core X and a GTX 1080 Ti is seeing the GPU show up in nvidia-smi at first, but it drops off after trying to use it for anything, sometimes just after having run nvidia-smi. Mattermost thread: https://chat.pop-os.org/pop-os/pl/1db8qbbhkb8n3pdrryx1r3bfcc