Ubuntu 20.04 RTX 3060 eGPU falls off the bus

I’m using Gainward Geforce RTX 3060 as an eGPU with Razer Core X. Connected to Dell XPS 7390 (2-in-1) running Ubuntu 20.04 and one external display through display port. Have tried switching the Thunderbolt 3 cable with no effect. I’ve had this setup for a couple of months now. Everything is new.

So - after approximately 30-60 minutes of using the computer, the keyboard, mouse, touchpad, and touchscreen stop working and I cannot interact with either the laptop display or the external monitor. If there is sound playing, it continues uninterrupted. Also my experience has been that the longer I use the computer during the day (that is, always shutting down and restarting), the time between the crashes gets shorter. At worst it’s been only a few minutes. There could be one light program running and it crashes nevertheless.

Looking at logs the GPU “falls off the bus”. I was able to get the bug report through SSH.

I’ve reinstalled Ubuntu multiple times, tried several different drivers, different settings, and the same problem occurs each time. There haven’t been crashes while playing games on Windows 10, which makes me suspect the Nvidia drivers on Linux are at fault - or maybe there’s some kind of wrong setting that causes this.

Thanks for any help!

data (2.1 MB)
bug report

Did you have that problem from the start? The cable seems to be fine, no pcie errors reported so must be something else.

I think the problem was there from the beginning, yes. But then, I used Pop Os for a while and had several other issues with the setup, so I can’t really trust my memory here.

Please try limiting clocks running
nvidia-smi -lgc 300,1750
to rule out power issues.

Tried your suggestion - at first almost 3 hours went by without interruption. Unfortunately now it’s back to a couple of minutes between crashes.

Could it be a hardware issue despite good performance on Windows?

The nvidia-smi setting is not sticky, does not survive a reboot. A difference between linux and windows driver is that boost clocks are unlocked in linux so peak power is higher,
Since it was not falling off the bus with that setting active, I guess it’s the psu of your egpu enclosing that can’t keep up with high boost clocks.

1 Like

That makes sense! Limited the clocks again and the system keeps running. Didn’t think this issue could be solved but it definitely seems like it :)

Thank you so much!

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.