Geforce 5090 Monster Hunter Wilds Crash After Shader Build

[LyingCak3]

NVIDIA Open GPU Kernel Modules Version

580.105.08

Please confirm this issue does not happen with the proprietary driver (of the same version). This issue tracker is only for bugs specific to the open kernel driver.

• [ ] I confirm that this does not happen with the proprietary driver package.

Operating System and Version

Arch Linux

Kernel Release

6.16.10-zen1-1-zen

Please confirm you are running a stable release kernel (e.g. not a -rc). We do not accept bug reports for unreleased kernels.

• [x] I am running on a stable kernel release.

Hardware: GPU

GPU 0: NVIDIA GeForce RTX 5090

Describe the bug

Monster Hunter Wilds crashes after shader build but only for Gefore 5090.

To Reproduce

Install Monster Hunter Wilds from steam. Use Ge-Proton10-25. Launch game Wait for shader optimization to complete Crash happens shortly after.

Bug Incidence

Always

nvidia-bug-report.log.gz

nvidia-bug-report.log.gz

More Info

No response

[Binary-Eater]

Thanks for the bug report. I just wanted to make sure if you are suggesting this bug only reproduces with blackwell and not with previous generation GPUs. We have a different shader compiler for blackwell that could explain the difference in experience with Monster Hunter Wilds.

[LyingCak3]

Correct, I have another machine with the same distro, kernel, and Nvidia open driver but a 4070 super instead and the game works without issue there.

[Binary-Eater]

I was taking a peek and it looks like we are tracking a potentially similar issue in bug 5547446 internally. However, I am not entirely sure the issue is identical, so I'll take a deeper look.

[konovalov-nk]

Attaching my own log as well!

nvidia-bug-report.log.gz

I'm running Proxmox + Arch Linux VM, with GPU+USB passthrough / IOMMU:

Host

• OS: Proxmox VE 8.4.1 x86_64
• Kernel: Linux 6.8.12-11-pve

Guest (Arch Linux)

• Arch Linux, kernel 6.17.8-arch1-1
• RTX 5090, 580.105.08 (open-dkms via pacman)

$ pacman -Qi nvidia-open-dkms
Name            : nvidia-open-dkms
Version         : 580.105.08-2

• Tried to run game via Proton Fix

symptoms/behavior: Shader build goes fine. The only problem I had is crash/black screen after shader build, where game didn't reach "Auto-save" modal window. I believe once game runs the auto-save, the shader file is then "validated/final". On subsequent runs with proper vkd3d-proton.cache the game loads it and compilation step goes from 0 to 100% in less than 5-10 seconds.

However, after the auto-save the game hangs after short burst of CPU/GPU utilization.

dmesg (on Arch guest) shows this:

[ 1631.656481] NVRM: GPU at PCI:0000:04:00: GPU-318496ea-c84f-7a8e-b9cd-6b997d4cb3f5
[ 1631.656486] NVRM: GPU Board Serial Number: 0
[ 1631.656487] NVRM: Xid (PCI:0000:04:00): 109, pid=8335, name=MonsterHunterWi, channel 0x0000000e, errorString CTX SWITCH TIMEOUT, Info 0x8c00a
[ 1779.677626] umip: MonsterHunterWi[17612] ip:15e5e61fa sp:417e98: SGDT instruction cannot be used by applications.
[ 1779.677631] umip: MonsterHunterWi[17612] ip:15e5e61fa sp:417e98: For now, expensive software emulation returns the result.

dmesg on Proxmox is clean.

Custom commands I used to launch the game:

WINEDLLOVERRIDES="dstorage,dstoragecore=n,b" VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/nvidia_icd.json  SDL_VIDEODRIVER=x11 gamescope --backend sdl -- %command%

WINEDLLOVERRIDES="dstorage,dstoragecore=n,b" VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/nvidia_icd.json %command%

VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/nvidia_icd.json %command%

All of them work and I have to specify VK_ICD because by default it would run on Arc310 (4GB GPU), which I'm using for HDMI outputs to 3 main displays.

Also attaching btop if helpful. You can see that game is actually loading into GPU memory and it starts working but then during load it goes 100% utilization and then dies out.

Thank you for working on this 🙇

For my next steps, I will try to downgrade to 575 and see if this would allow the game to run. Happy to assist with testing/running custom driver builds!

Update: after downgrading to 575.64.05 + installing Linux 6.12.58-1-lts the game launched fine! Haven't even observed vertex shader explosions / black textures yet but I only managed to go through first cutscene until the very first tutorial.

[LyingCak3]

Hello, I am wondering if there has been any progress on this issue? The game is still unplayable and this issue has been open for a while now.

[mtijanic]

The linked nvbug 5547446 is still open. Someone is looking into it, I imagine it'll show up in the changelog once fixed. This report is linked to it as well.

However, I'm going to close this issue here as this repo is about kernel modules only (and, theoretically, only issues specific to the open flavor of those) and userspace bugs are typically handled by other groups, and are supposed to be reported via [email protected] or the forums.

Now, I understand we can't expect people to know whether a given issue is a userspace or kernel module bug, so it's fine to open these issues; just be ready for it to be closed and redirected if it turns out you're in the wrong place :)

[Binary-Eater]

I changed the closed status just to indicate we did not actually solve the issue described. For folks wishing to engage with us on this issue, please open a Linux Graphics Nvidia forum post and do not hesitate to tag me there.

[Binary-Eater]

@LyingCak3 would you mind retesting with 590.48.01 and let us know if the game is working as expected?