protonscr

Cyberpunk 2077 fails to start when system contains nvidia dgpu and amd igpu

vkd3dclosed
HansKristian-Work/vkd3d-proton#2698 · opened 2025-11-11 by philipl · updated 2025-11-14 · 5 comments · github
Pphilipl 2025-11-11 github

I have observed on two separate systems that Cyberpunk fails to start if there is an nvidia dpgu and amd igpu present. This happens both when render offload is used and the amd igpu is primary and when the dgpu is primary and no render offload is configured.

The only way to get the game to work is to hide the igpu in the bios.

Software information

Cyberpunk 2077 - patch 2.31
No indication that any game settings are relevant as it crashes very early.

System information

  • GPU: nvidia 5090
  • iGPU: AMD MENDOCINO
  • Driver:
    • nvidia: 580.105.09
    • amd: mesa 25.2.6
  • Wine version: Proton Experimental 2025-11-07
  • VKD3D-Proton version: vkd3d-1.16-194-g7c0da1747a89 according to proton changelog

Log files

steam-1091500.log

BBlisto91 2025-11-11 github

Maybe a line of interest from the log? Not 100% sure.

free(): chunks in smallbin corrupted

vkd3d-proton haven't even loaded yet before the log ends. Dxvk have though.

Pphilipl 2025-11-12 github

So that points to the dxvk dxgi implementation? That's the only part of dxvk that would be involved? Should I file this as a dxvk bug?

Ddoitsujin maintainer 2025-11-14 github

I don't really see how our DXGI would break in only this game with only this particular setup given that pretty much all reasonably modern D3D12 games hit the same code paths inside DXGI. I mean, it could be, but it could also be ten thousad other things including just the game itself being broken.

Not much we can do without a way to reproduce anyway, the game works fine on my desktop which does have an Nvidia GPU paired with a dedicated AMD GPU.

BBlisto91 2025-11-14 github

Does removing lsfg-vk change anything?

Pphilipl 2025-11-14 github

Well, I booted up today to test whether removing Vulkan layers would make a difference, and suddenly everything worked. The main thing that changed was upgrading to mesa 25.2.7 yesterday. So I guess that's where the problem was. Maybe a mesa bug that got fixed, or maybe even a problem in how 25.2.6 was built (given the libc memory corruption error).

Of course, good news - with the crash out of the way, I get to file #2707. Sorry!

Proton versions