Linuxes don't like 2 GPUs & 2 Screens; Please help

[Stephen1R2]

Hello,

 

I have a weird issue on my main system which seems to bork any linux install I try. There is probably a magic boot/kernel parameter which would fix it but ???

 

My Hardware and the issue:

hardware ids:

0x1002:0x687f:1:5:Vega 10 XL/XT [Radeon RX Vega 56/64]

0x1002:0x67e3:1:5:Baffin Polaris11 [Radeon Pro WX 4100]

MB ASUS X570-p, with 3900X

Direct cpu  x16 slot: Vega 56 -- DP Primary monitor AOC 1440p ultrawide
0x1002:0x687f:1:5:Vega 10 XL/XT [Radeon RX Vega 56/64]
PCI Location (Bus:Dev:Fnc): 15:00:0

 

chipset x4 slot: R Pro wx4100 -- DVI secondary portrait screen
0x1002:0x67e3:1:5:Baffin Polaris11 [Radeon Pro WX 4100]
PCI Location (Bus:Dev:Fnc): 05:00:0

 

Windows 10 and now 11 work fine after a few hicups; working fine with the AMD downloaded auto-detector which installed the Radeon Pro driver package.

Ubuntu or pop-os (21.04) will install but only provide low res lvmw 1024x768 software render.

It turns out that the built-in amdgpu kernel driver is finding the wx4100 and not looking further perhaps because it has pci-e index 05:00:0 and the Vega is at 0F:00:0 for whatever reason. So the primary display, a 1440p ultrawide, is connected to an unclaimed adapter and so is software rendered.

I've found no documentation to force the use of both cards.

In manjaro-GNOME the live iso gets to the graphical-interface checkpoint and then a flickering of large print on the Primary display until I hit reset and give up.
on manjaro-KDE the live iso stalls/hangs at the graphical-interface checkpoint.

 

No... I cannot "swap" the cards. No I will not remove the wx4100 to make it work ( it does, but borks as soon as I slot it back in)

Any ideas for my training issues are also welcome; yes those are toothpicks, and there is no hotglue...yet

 

[Lightwreather]

If possible maybe try installing the Radeon Pro drivers? I'm fairly certain that has a different package compared to the one shipped ootb on Linux.

But it's been a hot minute since I've used Linux, so I could very well be wrong

[jde3]

You are trying to run one Xorg session that is running on two different devices? Yeah you may have issues with that out of the box. You need to setup a xinerama session.

 

To just get it installed you can turn one of the devices off in the kernel. (same way they do it in KVM PCIe Passthrough.)

[Wild Penquin]

15 hours ago, Stephen1R2 said:

It turns out that the built-in amdgpu kernel driver is finding the wx4100 and not looking further perhaps because it has pci-e index 05:00:0 and the Vega is at 0F:00:0 for whatever reason.

How do you know this (added emphasis to make my point clear)?

 

There is also a lot of other information, but I have no idea what you have based it on since there are no logs or other information.

 

Post the output of `lspci` and - most importantly - the Xorg.0.log, and perhaps even the Kernel log from a boot (`journalctl -kb`) (use pastebin or something for very long logs). Also, `ls -l /dev/dri` can be used to determine what GPUs the Kernel has found.

 

My guess would be that X.org does not handle several graphics cards too well per default. This might even be by design since a GPU could be used for many things (computing; separate X.org sessions) and unless the user tells it otherwise, it will choose only one to use? Most (all?) GUI monitor settings in DEs are not used to configure X.org that much but only the outputs (of the GPUs the X.org sessions has in use). There's information probably buried down somewhere in X.org documentation, but I guess little information online since running a single GUI with many graphic cards (in a non-hybrid graphics session with separate monitors on each) is a bit niché. Arch wiki has some examples on what to configure for multi-GPU setups (X.org is mostly uniform accross distributions so they should work in any, but check your distributions configuration examples). Anyways, I'd do the above steps first to determine what is going on.

 

Xinerama is an old, legacy way of handling multi-monitor setups and does not in any way help with several GPUs.

AMDGPU Pro has only advances for computing (OpenCL) users, for normal GUI users the open driver is better (but I could be wrong - doesn't hurt to try).

Edited March 23, 2022 by Wild Penquin

[YamiYukiSenpai]

Try installing with one GPU first, then plug the other one in.  Or remove the cable from one of the GPUs

 

I didn't have that issue with my RX 6900XT + GTX 1060 combo, IIRC, but I don't think I had anything plugged to the GTX 1060 at the time

[Master Disaster]

20 hours ago, Wild Penquin said:

My guess would be that X.org does not handle several graphics cards too well per default. This might even be by design since a GPU could be used for many things (computing; separate X.org sessions) and unless the user tells it otherwise, it will choose only one to use? Most (all?) GUI monitor settings in DEs are not used to configure X.org that much but only the outputs (of the GPUs the X.org sessions has in use). There's information probably buried down somewhere in X.org documentation, but I guess little information online since running a single GUI with many graphic cards (in a non-hybrid graphics session with separate monitors on each) is a bit niché. Arch wiki has some examples on what to configure for multi-GPU setups (X.org is mostly uniform accross distributions so they should work in any, but check your distributions configuration examples). Anyways, I'd do the above steps first to determine what is going on.

Can't be an X issue since on Manjaro with an AMD GPU the system will default to Wayland for both Gnome & KDE.

 

On 3/23/2022 at 4:48 AM, Stephen1R2 said:

Hello,

 

I have a weird issue on my main system which seems to bork any linux install I try. There is probably a magic boot/kernel parameter which would fix it but ???

 

My Hardware and the issue:

hardware ids:

0x1002:0x687f:1:5:Vega 10 XL/XT [Radeon RX Vega 56/64]

0x1002:0x67e3:1:5:Baffin Polaris11 [Radeon Pro WX 4100]

MB ASUS X570-p, with 3900X

Direct cpu  x16 slot: Vega 56 -- DP Primary monitor AOC 1440p ultrawide
0x1002:0x687f:1:5:Vega 10 XL/XT [Radeon RX Vega 56/64]
PCI Location (Bus:Dev:Fnc): 15:00:0

 

chipset x4 slot: R Pro wx4100 -- DVI secondary portrait screen
0x1002:0x67e3:1:5:Baffin Polaris11 [Radeon Pro WX 4100]
PCI Location (Bus:Dev:Fnc): 05:00:0

 

Windows 10 and now 11 work fine after a few hicups; working fine with the AMD downloaded auto-detector which installed the Radeon Pro driver package.

Ubuntu or pop-os (21.04) will install but only provide low res lvmw 1024x768 software render.

It turns out that the built-in amdgpu kernel driver is finding the wx4100 and not looking further perhaps because it has pci-e index 05:00:0 and the Vega is at 0F:00:0 for whatever reason. So the primary display, a 1440p ultrawide, is connected to an unclaimed adapter and so is software rendered.

I've found no documentation to force the use of both cards.

In manjaro-GNOME the live iso gets to the graphical-interface checkpoint and then a flickering of large print on the Primary display until I hit reset and give up.
on manjaro-KDE the live iso stalls/hangs at the graphical-interface checkpoint.

 

No... I cannot "swap" the cards. No I will not remove the wx4100 to make it work ( it does, but borks as soon as I slot it back in)

Any ideas for my training issues are also welcome; yes those are toothpicks, and there is no hotglue...yet

 

 

To be 100% sure you'd need to check the kernel logs from boot and see what the AMDGPU kernel modules reports.

 

sudo dmesg --human --kernel  --reltime

 

will get you a full kernel log in a nice readable format, hold page down till the bottom then go back up until you find the current boot and look for modules called AMDGPU.

 

You can also do

 

sudo journalctl -k -b0

 

to get the kernel logs for the current boot but its very hard to read since its just white text dumped onto a page.

 

Copy/paste them here please

[Stephen1R2]

Hello

 

Thank you for the advise. The issue was a little while back and I had given up (and nuked the install) on that system for linux since the X570 is supported with current Win11 and most things I do use that OS.

 

I had been curious since pop! OS with nvidia has been fully supporting my frankensteined (installed a Quadro M2000M) Dell M4700 that requires inf modding drivers (a bit of a pain) to run Win10 that I would like to keep working after Win10 EOL.

 

Two Workstations one an

ASRock Dual e5-2670v1 (R9-390)

Mint LTS success after enabling amdgpu disabling radeon in grub

so Yay! but Fedora WS failed to a blackscreen (can I text only an install?)

 

and one HP Z420 with EVGA-1070 that could be next on the Linux train.

 

But then I randomly downloaded an ubuntu 20.04 LTS, Ruffused it to usb, and it decided to "Just Work" on the Asus X570 when I made space for it on a spare nvme.

 

Full drag the window wherever on both screens so ????