Figuring out where in the display stack an eGPU + external 4k display is breaking
 

Hey all, I've been dealing with this issue for a long time and never really knew where to start diagnosing the issue, so would just appreciate any advice on where to hone in on the problem.

I have an eGPU and trying to connect my laptop through it to an external display over Thunderbolt.

Here's what I'm working with:

• Laptop: LG Gram 17 (i7-8565u, Intel UHD 620 integrated graphics)
• eGPU: Razer Core X
• Video card in eGPU: AMD RX 470
• external display: Sceptre 4k 60hz TV (connected via HDMI to GPU)
• OS: NixOS 21.05
• DE: KDE Plasma 5.21.3
• Graphics stack: X11
• Kernel: 5.11.14

So the egpu provides power over USB-PD, which does work fine when connected (it charges) and SHOULD automatically recognize the display through the GPU itself. Looking through the logs and output, it does indeed detect the graphics card just fine and even I think starts the driver.

Problem is it doesn't seem to detect the external display at all. No other displays show up in KDE display settings or in the /var/log/messages output. Other problems: if I originally booted up the laptop without the eGPU connected, when I connect the cable, the X session freezes up for some reason. Everything is still visible on-screen and I can move the cursor, but nothing is clickable and whatever was on the display at the time of connecting the egpu freezes completely (i.e. the clock in the screen corner doesn't change, UI elements such as the X to close windows don't light up when moused over etc..) until I either hard reboot/sysrq+b reset; ctrl+alt+backspace to kill the X session; or sometimes I think I can ctrl+alt+1 to switch to another virtual terminal.

Log up to this point (start = plugging in egpu; stop = when I reboot):
https://pastebin.com/bYUh2Fgd


If I connect the eGPU first and THEN boot up, the system is at least fully usable, although still no external display detected. But if I disconnect the eGPU and switch to using the laptop on battery while still booted, it either freezes immediately or sometimes keeps running fine until it sleeps or the display turns off, at which point waking it back up only shows a black screen with an unmoveable cursor. Then I have to reset again.

Log for that scenario (fresh boot while plugged in):
https://pastebin.com/5DGxSJzW

I know it's not exactly my hardware itself but probably something in the way it's configured with NixOS, because booting a regular Ubuntu live ISO works just great - hotplugging the egpu makes the display come up immediately and everything works smoothly. Although sometimes I have noticed the X server freezing thing even on Ubuntu too. Another small problem is I've never gotten either GNOME or KDE to run this 4k TV at 60hz - something that works fine on windows through the same egpu. But not as much of a priority as getting it working *at all* on nixos, and not forcing me to shut down & restart every time I unplug it.

I'll fully admit that I doubt I can lead you to a complete solution, but have a few ideas on 'poking about' that may or may not be useful:
and
looks like - does it at least show both GPUs as available, and can you force render-offload onto the eGPU? (I would just try glxgears for ease of use)


What happens if you try forcing an output via xrandr on the eGPU?

Finally just for general info, if you want to verify amdgpu is loaded (and maybe get more information in the process), use lshw (with suser/root permissions):
There should be a field titled 'configuration' that will show 'driver' and you should see amdgpu (which is, as far as I understand it, the only option for Polaris).


Sorry I have no ideas on the hotplug situation - I know I've looked into TB/eGPU things in the past (just to read about them) and it seems to be one area where drivers lag Windows (because there isn't first-party support in most cases, or so I've gathered), so some quirkiness seems to be par for the course.

Oh, and your links return 'this is a private paste' - you should be able to paste output with code tags here, or at least relevant portions of it though.

Thanks for the reply, and wow thank you - that actually does help!

Ok, awesome, didn't realize xrandr was indeed recognizing the GPU (and not just the kernel itself or...however that works). I'm not sure what exactly a "provider" is in xrandr vs. a screen/monitor/output though, still trying to make sense of all that.

That's really interesting too, it looks like the first display is the integrated graphics (judging by i915drmfb) and the second is..something, through the eGPU right? But then why is the "mode" on the first display listing the 4k resolution, does that mean the intel graphics are handling the external monitor/trying to drive it or something?


And as for the output of xrandr -q --verbose (for extra stuff like the hex identifiers):
[spoiler]
[/spoiler]

So I'm not exactly sure how to force the output onto the external monitor with xrandr, but on some quick googling it looks like the option: --setprovideroutputsource <prov-xid> <source-xid> is probably my best bet? But I'm not sure what to put as the source-xid.

Trying this:
so using "1" as the source-xid - is the only combination I've found that does anything at all, but it just kills my X session and dumps me back on plasma's login screen after a few seconds.
However - using --setprovideroutputsource 0xe5 0x47 (so the AMD graphics' provider hex ID as the 1st parameter, & the other provider as the 2nd) did work to make the external monitor show up in my plasma system settings menu! But it still didn't work even once I checked its "enabled" checkbox. I'm fumbling in the dark with this xrandr command though so probably doing that wrong.

And I just checked those pastebin links, they should definitely show up as public. Opened fine for me in an incognito window but I can paste here if you or anyone still can't see them

Glad it at least helped some. Something I would try next: can you connect the free-standing monitor directly to the laptop's IGP? (does it have its own HDMI/whatever output?) Does that work and get you both displays + everything is good so far? If yes, then I would connect the external GPU (do not hook the monitor up 'through' the external GPU - leave it on the laptop GPU), and try prepending the dri_prime command to some application and see if it runs. Example:

You will have to get the variable (in my example it is '1') from the --listproviders, it will probably be '1' as the 2nd GPU, which it is in your pasted output (but this could change if you add other devices/re-arrange hardware (not so possible with a laptop, but switch PCIe cards around in a desktop and those numbers will change)). I chose glxgears because its easy and will run on a lot of things, but you could send any 3D application there (for example you can set this as a 'launch option' in Steam for games, or prepend it to a command starting Super Tux Kart, or whatever).

If that all works, you'll have the RX 470 performing the 'work' while the Intel drives the displays (you would have to prepend this per-application, mind you) - probably not exactly what you originally set out to do, but it will hopefully at least get the RX 470 'working' for whatever heavy lifting I assume you've enlisted it for.