Qubes doesn't boot when the graphics card is installed

When we boot using the internal graphics card, the system starts up fine. However, after installing the 3060 Ti, the system fails to boot, and we can’t even access TTY2 to log in.

We do receive a signal through HDMI from the 3060 Ti at the LUKS password prompt, but nothing beyond that.

We also discovered that the Wi-Fi/Bluetooth module was defective, which caused the FLR error. To resolve this, we removed the module entirely.

Here’s what we’ve tried so far:

Adding rd.driver.blacklist=nouveau
Testing both the mapbs and noexitboot options (individually and together)
Adjusting various BIOS settings found online (we eventually restored defaults)
Removing the graphical LUKS interface and disabling quiet mode
Using an older Kernel



These images show the boot process with the 3060 Ti installed:



These images show the boot process without the 3060 Ti installed:


As a new User i cant upload files, so here is the content of the yml file:

Qubes-HCL-ASRock-B650M_Pro_RS-20241005-005338.yml

layout:
‘hcl’
type:
‘Desktop’
hvm:
‘yes’
iommu:
‘yes’
slat:
‘yes’
tpm:
‘2.0’
remap:
‘yes’
brand: |
ASRock
model: |
B650M Pro RS
bios: |
3.01
cpu: |
AMD Ryzen 7 7700X 8-Core Processor
cpu-short: |
FIXME
chipset: |
Advanced Micro Devices, Inc. [AMD] Device [1022:14d8]
chipset-short: |
FIXME
gpu: |
Advanced Micro Devices, Inc. [AMD/ATI] Raphael [1002:164e] (rev c3) (prog-if 00 [VGA controller])
gpu-short: |
FIXME
network: |
Realtek Semiconductor Co., Ltd. RTL8125 2.5GbE Controller [10ec:8125] (rev 05)
memory: |
64631
scsi: |

usb: |
4
certified:
‘no’
versions:

  • works:
    ‘FIXME:yes|no|partial’
    qubes: |
    R4.2.3
    xen: |
    4.17.4
    kernel: |
    6.6.48-1
    remark: |
    FIXME
    credit: |
    FIXAUTHOR
    link: |
    FIXLINK

Try to use kernel 6.6.42:

We tried all 3 Kernels that were available on the advanced boot options, down to the .38 (including .42) this did not make a difference.

Do you want to use NVIDIA GPU instead of integrated GPU in dom0? Or do you want to still use integrated GPU in dom0 and pasthrough NVIDIA GPU to some qube?
Just to note that GPU in dom0 is only used in dom0 and not used for graphics in the qubes so using better GPU won’t affect graphics performance in the qubes.
If you want to passthrough NVIDIA GPU to some qube and just want to be able boot with it connected to configure it later then you can try to boot with nouveau.modeset=0 rd.driver.blacklist=nouveau kernel command line options in GRUB.

The Endgoal is to use the GPU in a Gaming-Cube, thats why i have added
rd.driver.blacklist=nouveau to the grub parameters already.
i will add the nouveau.modeset=0 to that aswell and report later if that worked.

sadly it didn’t work. nouveau.modeset=0 didn’t do anything new.
we tried 6.6.48-1, 6.6.42-1 and 6.6.36-1.

Did you connect any display to your NVIDIA GPU?
Maybe there is a setting in the BIOS to select the primary GPU to be internal/dedicated?

We tried to boot with connecting the display to both the internal and the big GPU.

We didn’t find anything related to that in the bios.

Try to disable qubes autostart as well:

Along with disabling nouveau nouveau.modeset=0 rd.driver.blacklist=nouveau qubes.skip_autostart.

The underlying problem was that the inclusion of the graphic card changed the numbers of the pci devices, and that made the pci 05 device invalid as Ethernet-card.

qubes.skip_autostart => remove the faulty device, bind the correct one with the new number.

1 Like

Great that you solved your problem @Anti27 !

I am in a similar situation and, for the moment, my solution was to simply disable the NVIDIA (in the BIOS) and just work with the integrated graphic card.

Would you be so kind to write a set of detailed instructions that worked for you? Which files to edit, etc? Thanks!