My Issue
-
When playing games(such as Black Myth: Wukong and Stellar Blade) on CachyOS, my AMD GPU suddenly crashes. The display signal is lost completely, while audio continues to play normally. After the crash and then reboot, sometimes launching a game again immediately triggers the same issue.Based on the error log:
Apr 28 14:39:36 <hostname-redacted> kernel: amdgpu 0000:2d:00.0: device lost from bus!
Apr 28 14:39:36 <hostname-redacted> kernel: amdgpu 0000:2d:00.0: SMU: bus error for message: TransferTableSmu2Dram(18) response:0xFFFFFFFF
Apr 28 14:39:36 <hostname-redacted> kernel: in params:00000005
The GPU is no longer being recognized by the bus.
-
I have tried downgrading packages by rolling back to a snapshot, but the issue still persists.
-
I have come across similar issues, such as https://discuss.cachyos.org/t/screens-go-black-at-inconsistent-times-when-playing-games-amd-gpu/22438, which manifests similarly to what I am experiencing. However, I have tried the solutions mentioned there, and none of them worked.
Logs
2f25d2a
Solutions I Have Tried
- Reinstalled CachyOS, then ran the game directly.
- Roll back to a snapshot version that I vaguely recall being able to play stably at some point.
- Disabled hardware acceleration in Firefox.
- Set kernel parameters to disable power saving for amdgpu, such as
amdgpu.runpm=0, etc.
- Tried the CachyOS-LTS kernel.
- Switched Proton versions, from Proton-CachyOS to GE-Proton.
By The way, I tested the same scenario on Windows, and no issue occurred.
My Hardware Information
- GPU: AMD Radeon RX 9070 XT
- CPU: AMD Ryzen 5 5600X
- Motherboard: MSI B550M
- PSU: 1200W (confirmed stable)
Other Information
-
As mentioned above, I tested the same scenario on Windows and did not encounter the same issue. This rules out hardware problems, such as a faulty GPU.
-
The Steam launch parameters include: MANGOHUD=1, PROTON_FSR4_UPGRADE=1, game-performance.
-
Regarding the parameter amdgpu.ppfeaturemask=0xffff7fff in the cmdline of the error report, I already had this issue even before enabling it.
hello welcome to the forum
i just want to help, my current bootflag is like this, incase helps (you can ignore log related)
rw rootflags=subvol=/@ zswap.enabled=0 nowatchdog loglevel=7 systemd.show_status=true amdgpu.gpu_recovery=1 amdgpu.modeset=1 usbcore.autosuspend=-1 split_lock_detect=off initrd=\initramfs-linux-cachyos.img
amd overclock flag (managed by lact)
options amdgpu ppfeaturemask=0xFFFD7FFF
my current driver, kernel
mesa 2:26.0.5-4
lact 0.8.4-2.1
linux-cachyos 7.0.1-1
have you tried lact ? for controlling your 9070XT ?
what i did is change the performance level from Auto to Highest (so it’s stay within stock clock)
i didn’t play something heavy recently, only Endfield daily, and capped with 120fps lock via mangohud, sofar no problem on my end, never encounter crash (if i used auto, yes i do have crash as well in the past, around few month ago tho)
I have never used Lact to control my Radeon 9070 XT. The kernel parameter amdgpu.ppfeaturemask=0xffff7fff in cmdline was added after asking ChatGPT for possible solutions. Before adding this parameter (and now I know it’s a parameter for enabling overclocking), I still had this issue. Perhaps I could try using Lact.
Thank you for your reply.
yeah F7 was the old featuremask, i think last month ? lact suggested new featuremas code because of new version of mesa i guess, so i’m using D7 now
as for lact, i did it for undervolt and control the auto overclock on my card
After cleaning the gold fingers of the graphics card and reseating it, I have not encountered this issue for now.
what an unexpected plot twist huh ?
well atleast you managed to fix it, enjoy ~
Well, the issue just appeared again. It’s not solved.
Here is the latest log report.
can i see your LACT config ? undervolting etc (if you use it to tame the incorrect clock speed report with auto overclock mode, atleast on my end)
also i seen lot of kwin_wayland[1234]: atomic commit failed: 无效的参数 (device busy)
for the sake of my curiousity, can you add this too on your boot param ?
amdgpu.gpu_recovery=1 amdgpu.modeset=1 split_lock_detect=off
This is my LACT config. I do not have overclocking enabled.
I added the boot parameters as you suggested. Let me see how it goes from here.
tldr: try change your performance level in lact to highest clocks too, and set -500 offset for gpu clock for safety purpose so it won’t breach your stock clock
i want to see the bottome ones, also performance level don’t set to automatic, set to highest clock for safety purpose (to make the “STOCK” clock as the highest)
that’s the problem for my end as well, if i use automatice performance level, my clock speed easily reach 3.2ghz LMAO and yes it will crash my desktop
so what i did in my previous post (2nd post i think)
i set the performance to highest clock
and just in case, use -500 offset, raise a bit of my vram clock (you can ignore this) and undervolting (gpu voltage)
if you set performance level to manual, it will apply the clockspeed settings in below section (and always go -500 offset for starter) so your gpu clock is near the stock level, then you can raise it bit by bit if you want to OC your gpu clock
also good luck on this attempt, if it’s the same problem like mine (gpu clock is too high, due automatic boost), then what i share to you should fix your issue
After add that to my boot param, it seems there isn’t much change in the error logs.
The settings at the bottom of LACT are as follows.
After enabling the overclocking settings, the defaults are as follows:
Here are the settings after I modified them. I only adjusted the frequency, because when I used AMD Software to lower the GPU voltage on the Windows platform, it often caused the settings to revert to their defaults.Maybe I will try slowly lowering the voltage later. Based on my experience, this should work, because when I played games that didn’t fully utilize the graphics card a few days ago, I didn’t encounter this issue.
yeah, if the issue only triggered by full load on gpu
seems same problem as mine, which is gpu clock is too high (even from stock)
so hopely with performance level at highest clocks, and -500 offset should suffice to mitigate this issue
as for the undervolting, you can try bit by bit, as usual in OC guidebook ~
my value is stable around -80 ~ -90
also another trick to restrict your gpu is, to set fps limit
i’m mostly using mangohud with 120fps caps (in case helps, reduce power draw as well)
my gaming profile for my lact should be like this
So far, things are looking good. If there are any further developments, I will update the thread.
I first tried underclocking, and then I tried not underclocking but instead lowered the voltage by 50. I’ve been using it since then, and so far no issues have appeared. Generally speaking, undervolting shouldn’t make the graphics card more stable. Although it’s strange, it has indeed been fine. For users experiencing similar issues, you can try a similar approach to mine.
If the problem occurs again, I will continue to update the thread.
Hey i faced the same issues as you so far what worked for mt 6800XT was to increase the Minimum Gpu Clock and gaming in cachyos-ltx instead of the normal kernel.
Before this my base gpu clock was 500 and i think maybe the big boosts caused it to spike and make in unstable in linux as opposed to it being fine on Windows.
yeah seems the kernel is incorrectly reading the GPU frequency, so better tune in with lact
yes it should make your system unstable (if it’s too much undervolting) normal value is quite stable for AMD gpu atleast
and it seems -50 for you was a breeze, congrats
my threshold for undervolting is -95 for now (with latest mesa)
Unfortunately, although I no longer have the issue on CachyOS, the same problem has appeared on Windows 10. Even after reinstalling the system, the issue persists on Windows 11, whereas it did not occur before. I tested my graphics card by installing it in my friend’s computer, which ruled out the graphics card as the cause. The problem should now be focused on my motherboard and CPU.
This issue is most likely no longer related to CachyOS. Maybe I’ll find a computer repair technician to figure out which piece of hardware is actually faulty. This thread is probably concluded.
