No devices were found with Ubuntu 20.04 + RTX 3060 Ti eGPU setup

[drecate]

I’m trying an eGPU setup with a Thinkpad X1 Carbon Gen 10 laptop running Ubuntu 20.04 (Linux kernel version 5.14.0-1059-oem). I install the nvidia-driver-530 driver (also tried versions 525, 515, 510, and 470) using the Additional Drivers menu. After rebooting, running nvidia-smi gives No devices were found.

Running lsmod | grep -i "nvidia" gives

nvidia_uvm           1277952  0
nvidia_drm             65536  0
nvidia_modeset       1273856  1 nvidia_drm
nvidia              55717888  14 nvidia_uvm,nvidia_modeset
drm_kms_helper        258048  2 nvidia_drm,i915
drm                   557056  16 drm_kms_helper,nvidia,nvidia_drm,i915,ttm

Running lspci | grep VGA gives

00:02.0 VGA compatible controller: Intel Corporation Device 46a6 (rev 0c)
22:00.0 VGA compatible controller: NVIDIA Corporation Device 2489 (rev a1)

Running nvidia-bug-report.sh gives, among other things, Failed to initialize the NVIDIA GPU at PCI:34:0:0.

I’m aware of workrounds for Proxmox (i.e., echo 1 > /sys/bus/pci/devices/0000\:0X\:00.0/remove,
echo 1 > /sys/bus/pci/rescan) as well as kernel parameters such as rcutree.rcu_idle_gp_delay=1, but nothing I’ve tried works. I know that the problem is not with the GPU or the enclosure, since they work fine with another laptop.

nvidia-bug-report.log.gz (333.6 KB)

[drecate]

Here is the solution/hack(?) that worked for me:

1. use nvidia-driver-515-open; it has to be the open source driver, and the version also seems to matter (525 and 530 seemed to cause XID 79 fallen off the bus error when the GPU is even slightly stressed).
2. add pcie_aspm=off nouveau.modeset=0 nvidia.NVreg_OpenRmEnableUnsupportedGpus=1 to the kernel option.

This is based off this thread: [SOLVED] NVIDIA drivers failing to load eGPU on Ubuntu 22.04.1 BIOS 3.06 beta - #17 by Ilya1 - Community Support - Framework Community

It’d be great if any expert can help elucidate why these settings in particular work.