Path of Exile 2 shuts down my machine
Most likely to occur with Vulkan. Seems isolated to this title. Running latest stable HTPC image.
I assumed initially this was hardware instability but I’m not as sure now:
- CPU: 13600K. I disabled all MSI “Enhanced Turbo” settings. I reseated the cooler. I swapped the cooler fans out (I used Noctua A12 LS, swapping in regular A12 later today). I checked the Lite Load mode and set it to 5. I set strict power limits: 125W long, 181W short.
- RAM: GSkill Ripjaws S5, 64GB, 6000MHz. I tried bringing the clock down to stock and retraining. No dice.
- GPU: Sapphire Pulse 7900XTX. Newly returned after an RMA. I assumed this may be a transient spike, but… it happens even on low graphics settings with Ultra Perf scaling. Setting a 60FPS cap seems to make it less frequent, but not unlikely.
- PSU: SF750. I have an SF1000 lying around that I can swap in, but I was planning on returning that.
The reason I’m not so sure that this is a hardware issue: I regain stability after deleting the shaders. I’m also entirely stable when playing Cyberpunk and other demanding titles.
55 Replies
Ran a 12 hour prime95 torture blend test. No issues detected. Temp never broke 68C even though I’m using a U12S and A12-LS instead of full speed A12! Unlikely to be Intel degradation
I've actually had this bug once as well
I have a feeling that putting a frame limit on the game with mangohud would fix it
But I haven't had a chance to test yet, my theory is that it has an unlocked frame rate at certain times that can cause the GPU to render way too many frames and hit a power limit
If I figure out exactly what it is I will let you know
This sounds really plausible to me fwiw
I just FurMarked the GPU to really check it’s operating normally (it is fully within expected bounds). Highest draw there was 305W (1s refresh of sensors), but I probably didn’t catch any transients. I can’t imagine it coming close to tripping OPP on my PSU (~950W).
Will test with:
MANGOHUD_CONFIG=fps_limit=120 mangohud %command%
Hard no dice
As an alternative:
AMD_DEBUG=sync_to_vblank %command%
Just out of curiosity, if you change the rendering engine from Vulkan to DX12 does it work?
You can change it with an
ini file. Don't remember which one, but protondb has a bunch of NVidia users that do that to play the gameHard crashes still occurred previously with DX12.
What about DX11?
Haven’t tried dx11. I can give that a shot momentarily. So far so good with the sync_to_vblank debug command
Oh nice!
that does seem to imply there’s something up with the game’s frame limiter / vsync
Aaaand hard crash.
You know, I've wondered that as well. I had issues with specifically this game where locking my screen or switching ttys would crank up my thermals (max them out), but without mangohud, it doesn't do that. My guess was mangohud's frame limiter somehow messing with the games one
This is wild though bc at the time of hard crash…
Nothing was near 100% usage
GPU: 65%.
CPU: 14%
Eh, I wouldn't think a thermal issue would do it, since the hardware should throttle not crash
What's your journalctl look like?
Right but at such low usage I can’t imagine a transient triggering it either
Use -k flag to get kernel stuff
Idk what transient means 😂
I have noticed on my device tho, that a GPU reset occurs when using DisplayPort
I wonder if you're seeing different symptoms for the same issue
Are you on KDE?
I'm using the latest HPTC image, no changes
Piped output of
journalctl -kthis is the current boot though, not last, right?
Does the game hard crash your PC while you play the game / enter loading screens?
Yeah, add
-b -1Also, it looks like you're using gamescope?
def using gamescope, i believe that's the default bazzite-htpc behavior
Crashes tend to occur while standing in town
What's htpc? Not sure what that is actually
The game has current issues with some CPUs and multithreading, could that be related?
All of these errors in that log wow
The logs definitely have errors, several of them AMD related, but I think the other ones seem more related
Oh huh
Are you plugged in with DP?
DisplayPort => HDMI cable
Should I just go direct with HDMI?
looks like all those errors are from the DP-4 connection somehow
Try that yeah
Weird. Okay, will swap connections, reboot, and see what happens.
Are you on KDE? Idk what desktop env your version comes with
KDE, yes
Good to know. If the HDMI thing works, then you might try to see if this fixes the DP. The errors don't look the same, but it is a Mesa bug specifically to do with DP so who knows: https://discord.com/channels/1072614816579063828/1236076247851733124/1319769115984003124
(isnt framework specific)
Your note here is fascinating:
AMD Mesa has a bug with their direct scannout feature on the GPU. Iirc, this only occurs if you're using the DP-USBC port on the back of the GPU, so using one of your side modules for HDMI or DP might also work around it for nowThis would 100% occur on all DP -> HDMI setups since that operates over DP-USBC
Not quite. On Framework, the DP-hdmi is though the motherboard, meaning the MUX is involved. I don't really understand much of the hardware past that tho 😅
But the back of my laptop is direct DP like your one is
Just booting into the game now and testing with HDMI… fingers crossed.
🤞
You're right tho, I probly should be more clear, since yeah all DP-HDMI setups use that protocol 😅
As an aside, 120Hz (and VRR) are ROUGH with hdmi output
No crashes though but I think this lends support to your thought here
Bc iirc — 4K@120 needs HDMI 2.1 which AMD cannot implement
Swapping back to DP, adding arg, testing again
Immediate crash
Damn
A very different log
This does include the test with direct HDMI prior to swapping
Trying this once more HDMI => DP, with the Display set to standard operation (60Hz / no HDR / no VRR — “Game Mode” disabled) and without the flag
Not tech savvy but found this https://www.pathofexile.com/forum/view-thread/3378857 two people mentioned playing the game with Proton 7 worked for them
Path of Exile
Technical Support - Game crashes loading towns (Linux) - Forum - Pa...
Path of Exile is a free online-only action RPG under development by Grinding Gear Games in New Zealand.
4:10:16 bazzite kernel: amdgpu 0000:03:00.0: [drm] REG_WAIT timeout 1us * 100 tries - dcn32_program_compbuf_size
That seems to be the most likely candidate for the issue
My thoughts too
Means it's some kind of mesa bug, it's possible it's been reported
Confirmed it’s 100% fine at 60Hz
Iirc we’re a minor version + a few patches behind on mesa?
Maybe fixed there 🤔
Even @ unlimited frame rate — where the game engine is pumping out 120+
So uh, seems potentially not related to frame limits — but something to do with the classically annoying 4K@120 nonsense. One of these days… maybe… the lawyers will step out of the way and let this work ootb in HDMI
I mean using DP should've worked then, no?
🤔 not unless the issue is DP-USBC, right?
It’s plugged into a tv — which only takes HDMI inputs
It doesn’t seem this has been reported before searching through their Gitlab. In a way, that’s kind of exciting.
Ooohhh
Unfortunately can’t open the issue: between never needing GitLab (and working at their competitor…) I never made an account and they’ve limited new accounts to be unable to create issues
Managed to trigger this in Spider-Man remastered:
- I attempted to turn on Ray Tracing in 120Hz. Same log produced as POE, but crashed the game instead of the machine.
- Then, launched with SteamDeck=0. hard crash (reboot) being triggered by that same line, without ray tracing or anything like that.
GitHub
linux/drivers/gpu/drm/amd/display/dc/hubbub/dcn32/dcn32_hubbub.c at...
Linux kernel source tree. Contribute to torvalds/linux development by creating an account on GitHub.
the exact line causing the crash
Oddly: hasn’t been updated in 8 months. Could well be downstream but I’m stumped.
Like it could be RADV. It’s unlikely to be DXVK/VKD3D-Proton given PoE is running on Vulkan (I assume Spider-Man is DX12)
Forced an update to all latest packages via Discover. This is where my knowledge of the Linux fs starts to fail me: driver updates and package updates done this way (say mesa etc) won’t apply, right?
So far no crash but I’m genuinely surprised by that?
Correct, steam is a system package so it doesn't use any of those flatpaks
Installed the SF1000. No crashes so far. The sf750 PSU passed a multimeter probe… but is definitely defective. It hits OCP at 650W: below its rated power headroom.
This machine is cursed istg.
Gpu rma. Intel 13th gen RMA. PSU defective.
At least my ram and board are fine.