protonscr

[amdgpu] UE5 & DX12 always crashes: Process GameThread pid 24526 thread vkd3d_queue pid 24578

vkd3dclosed
HansKristian-Work/vkd3d-proton#3139 · opened 2026-07-01 by arrowgent · updated 2026-07-21 · 4 comments · github
Aarrowgent 2026-07-01 github

Please describe your issue as accurately as possible. Include screenshots or videos if relevant
for several months of testing UE5, proton 10, proton 11, FSR4, and tweaking LACT and grub kernel configs, i have been trying to narrow down games crashing.
i can easily run a benchmark for hours
i can run most unity games at 4K60

as soon as i start a UE5 game on proton 10, 11, experimental, or proton-cachyos
the game will eventually crash between 1-30 minutes
sometimes immediately as the menu launches
sometimes waiting and watching the LACT or AMDGPU_TOP

i can do no further than narrow down that vkd3d-proton is causing an issue
this has been on going since kernel 6.14, mesa 24.x, proton 10
if it is mesa as many reports suggest on gitlab issues
or AMD firmware or amdgpu kernel! WHO KNOWS ANYMORE
i have even tested with nerfing the SCLK and PWR using LACT (BELOW STOCK CLOCKS)
my thermals are within spec, GPU never hits above 90C. CPU maintains throttling with edge_junction at or below 100C --- stupid hot laptop. i can somewhat control temperatures by power saving or reducing PWR etc. this does not prevent crashes and is not the cause (not overheating)

Software information

Name of the game, settings used etc.
unreal engine 5, DX12 -- vkd3d-proton
many games, here is one:
Gothic Remake (DEMO)

System information

AMD Ryzen 9 7945HX
GPU AMD Radeon RX 7900M navi31
Mesa 26.1.3 - kisak-mesa
Vulkan 1.4.350
Linux 7.1.2-amdr18
gcc (Ubuntu 15.2.0-16ubuntu1) 15.2.0
Ubuntu 26.04 LTS
  • GPU:AMD Radeon RX 7900M navi31 (RDNA3)
  • Driver:26.1.3 - kisak-mesa
  • Wine version: proton 10.0-4, proton 11+, proton experimental, proton-cachyos-11.0-20260602-slr-x86_64_v3, experimental-bleeding-edge-11.0-389079-20260701-p918c96-w9f7ab6-deea51b-v423207
  • VKD3D-Proton version:
    1213.597:0138:013c:info:vkd3d-proton:vkd3d_get_vk_version: vkd3d-proton - applicationVersion: 3.1.0.
    1213.597:0138:013c:info:vkd3d-proton:vkd3d_instance_init: vkd3d-proton - build: 4232071c346c0a7.

Log files

Please attach Proton or Wine logs as a text file:

  • When using Proton, set the Steam launch options for your game to PROTON_LOG=1 %command% and attach the corresponding steam-xxxxx.log file in your home directory.
  • When using regular Wine, use wine game.exe > game.log 2>&1 and attach the resulting game.log file.
    dmesg.txt & gothic proton_log=1

steam-3448280.log

amd_ring_dmesg.txt

today i tweaked again trying to find ANYTHING the grub kernel config options:
GRUB_CMDLINE_LINUX_DEFAULT="plymouth.debug plymouth.enable=0 systemd.show_status=1 ipv6.disable=1 preempt=full threadirqs amd_pstate=guided pcie_aspm=off amdgpu.gfx_off=0 amdgpu.ignore_min_pcap=1 amdgpu.gpu_recovery=1 amdgpu.lockup_timeout=10000 iommu=pt amdgpu.dcdebugmask=0x10 amdgpu.ppfeaturemask=0xfffd7fff"

i can go and use the default base repo Kernel with no kernel config options
these games will still crash

Aarrowgent 2026-07-02 github

https://github.com/HansKristian-Work/vkd3d-proton/issues/3004

ill just say this is a duplicate
closed

again noted:
LACT (and firmware) reports the RDNA3 PWR & max boost clocks incorrectly. setting these lower (to stock) or even BASE clock does not change the crash
neither does disabling overclock or disabling LACT etc.

& Thermals are within spec

can run FURMARK.x86_64 for hours. & while live tweaking PWR+SCLK to observe changes.

Aarrowgent 2026-07-03 github

i went back and "disabled all changes" down to a basic installation
(note this has been ongoing since 2024)

no changes in behavior.
4K60 on display:1 (2nd monitor) with any other game has no issues at high temp 99%-100% gpu usage etc.

  • open any UE5 DX12 game
    problem resumes, crashes within 1-30 minutes

tried disabling FSR/Upscaling etc.

i give up

Ttyrypyrking 2026-07-14 github

@arrowgent This is an issue with RX 7900 AMD series mesa driver specifically. A semi-decent fix is already live for a long time on mesa-git in AUR - newest 26.2.0 is what you need. I'm already working on a full fix that will resolve the rest of the issues introduced with the shaky driver workarounds.

Aarrowgent 2026-07-17 github

interesting...

i checked memtest_vulkan and discovered the VRMEM hitting 99C might be causing part of my issue
where amd shuts down the SCLK & MCLK for thermal protection
however, no dmesg/journal is reported when this happens

but yes, its known that amd firmware has some big issues aswell

thanks for the note

-- edit
its also troubling having spent over a year searching amd/drm gitlab for mesa and kernel for specific issues that are all over the place for RDNA3
ive tried everything. basically its impossible to find a "fix" that someone already reported with a workaround
so its going to be one of several things:

  • software issue
  • hardware issue
  • thermals

im on mesa 26.1.5 currently, and i just avoid using UE5 because it always give me issues...

Proton versions

Launch options

Upstream links