BC-250 (RADV GFX1013): GPU hang, then full system lockup ~50 s into the opening, right after the light (stock clocks) - same pattern as #228

Open
  • Is something wrong?
#527 2 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
52/100
Issue type
Bug
Clarity
Clearly specified
Activity status
Active
Tech stack
c, linux
Domain
backend, performance

Research direction

Start by comparing the attached savagewind.log and kernel logs with issue #228, which reports a similar RADV GPU hang fixed by disabling graphics pipeline libraries. Check how 0.4.1 handles GPL settings and GPU-specific defaults; the report says the attempted environment switches did not change the startup log. Done means identifying a reproducible way to disable GPL or confirming whether GFX1013 should default to GPL off.

Written by the indexing model from the issue text.

Description

bug
Before you start
Savage Wind version

0.4.1-alpha.4 (latest)

Game version

1.9.0

Your PC

SteamOS (Linux 6.16.12-valve24.5, KDE Plasma 6.4.3, Wayland), AMD BC-250 (RADV GFX1013, Mesa 26.0.0-devel dev snapshot), 12 CPUs (stock 6c/12t), 14 GiB RAM

Where in the game

Opening in the Shrine of Resurrection, right after the light (Great Plateau start)

What happened, and how to make it happen again

The game starts and plays the opening at 60 fps. About 50 seconds in, after the light, the whole machine locks up and needs a hard power-off (no ping, no SSH). Every time I tried (at least 3 times, 2 of them at stock settings with the kernel log captured), in desktop mode and in SteamOS game mode. I had earlier used the BC-250 toolkit overclock, with occasional lockups in other games, but I reverted it fully (stock clocks, stock core count, stock memory split) and the hang still happens.

Related: #228 describes the same pattern on an RX 9070 XT / RADV: hang ~46 s in, "ring gfx_0.0.0 timeout" naming savagewind, right after "[nvn] SetDepthClamp on". My log ends the same way. There it was fixed by turning off graphics pipeline libraries (GPL). My log shows "graphics pipeline libraries on". The difference: on the BC-250 the failed GPU reset takes the whole machine down.

Kernel log, captured live over SSH on two separate crashes (the disk journal loses it). Both show the same sequence. The attached savagewind.log is from the second one (04:14, with GPU_OVERLAP=0 set, which had no visible effect):
04:03: ring gfx_0.0.0 timeout, signaled seq=20255, emitted seq=20259 (savagewind pid 7238)
04:14: ring gfx_0.0.0 timeout, signaled seq=30160, emitted seq=30164 (savagewind pid 7241)
Then, both times: Starting gfx_0.0.0 ring reset / Ring gfx_0.0.0 reset failed / GPU reset begin! / MODE1 reset / PSP mode1 reset / VRAM is lost due to GPU reset! / PSP is resuming...
(the log ends there; the host is unreachable afterwards)

Last lines of savagewind.log before the hang (the log may lag a few lines):
[place] swr:1 x=-1130.00 y=237.40 z=1914.50 hour=5.00
[nvn] SetDepthClamp on (depth clamp supported)
[shadow] format 70 under a shadow sampler: depth comparison not supported
[shadow] a 448x448x3 R16_UNORM image sampled with depth comparison: through a D16 copy
[blur] distance blur's half-size read made depth-aware
[post] scene 1280x720 graded -> 1280x720 (format 64)

Also in savagewind.log: 24 of 21464 shaders failed to translate (compute shaders from agl_technique_pfx, and 4 fragment shaders "tlds info 6").

Switches tried: SAVAGEWIND_GPU_OVERLAP=0 and SAVAGEWIND_VK_NO_FEEDBACK_LOOP=1 (both confirmed in /proc/PID/environ), plus SAVAGEWIND_VK_NO_GPL=1 and SAVAGEWIND_VK_GPL=0. In every case the startup log still prints "GPU overlap on", "attachment feedback loops on" and "graphics pipeline libraries on". I couldn't find a way to turn GPL off in 0.4.1, so I couldn't test whether GPL is the cause here.

Questions: How can GPL be forced off on 0.4.1? Could GFX1013 be added to the GPUs where GPL is off by default (as for RDNA 4 + RADV)?

amdgpu-hang.txt

amdgpu-hang-A.txt

vulkaninfo.txt

Settings

60 fps, 1280x720 windowed, VSync mailbox, no upscaler (read from the log)

savagewind.log

savagewind-A.log

Before sending
  • This report contains no game files, keys, firmware, generated game code, save files or links to copies of the game.
Dominant language
C
Stars
811
Forks
37
Avg merge
17m
Merged PRs (30d)
13

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from Revivedx/SavageWindRecomp

All issues in Revivedx/SavageWindRecomp

Similar issues

More C issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.