The Nvidia driver stopped working on Linux for hybrid graphics cards starting with kernel 6.15 onwards

Graphics / Linux Linux
, ,
1

My computer is a Dell G3 3590, with a 10th generation Intel i5 processor, 12GB DDR4 RAM, and an Nvidia GTX 1050 3GB.

I’ve been using Linux on it for several years without any problems, but now with driver 580 I’m having issues. I’ve noticed, talking to other people, that this only happens with hybrid GTX 1050 cards; the problem doesn’t occur with non-hybrid cards.

If I try to use kernel versions 6.15 or higher, like 6.16 and 6.17, the Nvidia driver fails to recognize the card; it’s as if it doesn’t exist. Looking at dmesg, I found the following output:

[ 7.409460] nvidia 0000:01:00.0: Unable to change power state from D3cold to D0, device inaccessible

Adding the flag pci_aspm=off to grub, the driver recognizes the Nvidia card in nvidia-smi, but even forcing it, no app is rendered by it.

I already tested it on Debian unstable, Fedora 43, and Biglinux (based on Manjaro), all presented the same problem, with the difference that in Debian unstable the driver is not version 580.

2

Hi @mozaga.iarle, welcome to the NVIDIA developer forums.

Thank you for bringing this to our attention.

I moved the post to the dedicated Linux category and tagged the right people to get better tracking.

Thanks!

4

Hi, do you know if there is any update about this issue in some other post maybe? things are so silent here

6

Not that I am aware of. It might take a few days, especially since there was a public holiday in the US yesterday.

7

Hi @mozaga.iarle , thanks for reporting this issue. Could you please help to capture a bug report using sudo nvidia-bug-report.sh once you see this error message and upload it here? Thanks!

8

nvidia-bug-report.log.gz (337.6 KB)

Thank you! Here is the output from sudo nvidia-bug-report.sh

9

Hi @mozaga.iarle , sorry for the delay.
This issue could be due to a loose connection.
Could you try reseating the GPU to check if that is the issue?
Also, could you try installing an older driver version eg 570 and see if the issue is still seen? Thanks.

10

Before move to Archlinux I tried in Debian 13 with the driver 555 and had the same issue in these new kernel. I also can Change here to Kernel LTS 6.12 and post the bug report. Maybe this can help you also?

(about loose connection, its a laptop GPU, the plate is soldered)

11

nvidia-bug-report.log.gz (831.8 KB)

Here is the output from sudo nvidia-bug-report.sh

Using Kernel archlinux 6.12.61-1-lts and driver 580. Maybe this can also help to see some differences why the driver works in lts and not in newest kernel? (6.14, 6.15, 6.16, 6.17+)

12

Thanks @mozaga.iarle.
I see that the driver gets loaded properly in the new bug report that you uploaded.
Could you provide a bug report with pci_aspm=offon 580 and new kernel where you see GPU falling off the bus?

13

Yes, as I said before, its working only with kernel 6.12 LTS

Sure, I’ll add pci_aspm=off + newest kernel and send the bug report

14

nvidia-bug-report.log.gz (788.5 KB)

Here is the output using
nvidia driver 580
kernel archlinux 6.17.9
pcie_aspm=off

Why in LTS Kernel 6.12 its working with no flags and in the newests kernel not? this is the big question 🥲