RX 7800 XT (PowerColor Hellhound) — recurring GPU hang across nearly every driver version tested, on Windows AND Linux

Hi,

I’ve spent several weeks diagnosing a recurring GPU hang on a PowerColor RX 7800 XT Hellhound (dual BIOS). Posting the full dataset here — hoping either someone recognizes this exact signature, or an AMD rep can use this for the driver team.

SYSTEM
GPU: PowerColor RX 7800 XT Hellhound (dual BIOS switch, both OC and Silent profiles tested)
Motherboard: Gigabyte B550 Aorus Elite V2
Monitor: ASUS TUF Gaming VG27WQ, 165Hz, FreeSync Premium (VRR disabled throughout testing; both DisplayPort and HDMI tested)
OS: Windows 11 and Linux, dual boot, same physical hardware

DRIVER TESTING
I’ve tested nearly every driver version across a 2.5-year span (31.0.24027.1012 / Adrenalin 24.3.1 from March 2024, through 26.2.2, 26.7.1, 25.12.1, and others in between). The issue is present on every single one, but the pattern differs depending on version:

  • On the oldest driver (24.3.1), games run completely stable — zero crashes, even in long sessions. The GPU hang only occurs at idle (see symptom below).
  • On current/recent drivers, the same GPU hang additionally occurs during active gameplay (one confirmed case: CryEngine title, “D3D12 Device Removed”, shortly after raising in-game graphics settings), on top of the idle issue still being present.

SYMPTOM (idle, present on all driver versions)
After the system sits idle for several hours, any light GPU activity (opening a browser, playing a video) causes the local display to freeze, then go black for 2-4 minutes before self-recovering. Never occurs under sustained heavy gaming/compute load (tested for days at a time, GPU temps never exceed 65C under 100% load, no overclock/undervolt). During one freeze, the system remained fully reachable via Chrome Remote Desktop from another device while the local monitor stayed black — confirming the hang is isolated to the GPU/display subsystem, not a full system hang.

WATCHDOG DUMP HISTORY (C:\Windows\LiveKernelReports\WATCHDOG, parsed from DUMP_HEADER64: BugCheckCode at offset 0x38, params at 0x40-0x58)

Incident 1 — 2026-08-03, 23:01-23:07 (driver 31.0.24027.1012, idle): 0x141 → 0x117 → 0x193 → 0x193 → 0x1B0
Incident 2 — 2026-08-13, 23:10-23:15 (same driver, after BIOS profile switch + ASPM/MPO/ULPS tweaks, idle): 0x141 → 0x141 → 0x1A8 → 0x193 → 0x1B0
Incident 3 — 2026-08-15, 14:14-14:19 (driver updated to 26.2.2 via clean DDU reinstall, idle): 0x141 → 0x117 → 0x1B0
Incident 4 — 2026-08-15, 23:08 (26.2.2, during active CryEngine gameplay, D3D12 Device Removed): 0x141
Incident 5 — 2026-08-22, 18:34-18:35 (26.7.1, driver-only install, crashed on first reboot after install): 0x141 → 0x141

All incidents show driver module amdkmdag.sys loaded (standard WDDM KMD).

LINUX — INDEPENDENT REPRODUCTION (journalctl -k, amdgpu, same physical card, before any Windows-side tweaks):
SMU: No response msg_reg: 29 resp_reg: 0
amdgpu: Failed to disable gfxoff!
amdgpu: ring sdma0 timeout, Ring sdma0 reset failed
amdgpu: Failed to terminate ras ta
amdgpu: suspend of IP block failed -22
amdgpu: resume of IP block failed -62
amdgpu: GPU pre asic reset failed
amdgpu: GPU Recovery Failed: -5
Multiple kworker threads (ttm_bo_delayed_delete) blocked 122+ seconds waiting on GPU fence during this event.

WINDBG !analyze -v ON THE 0x1B0 DUMP (Incident 2, 23:15):
VIDEO_MINIPORT_FAILED_LIVEDUMP (1b0)
Arg1: 0000000000000001, Add device failed
Arg2: ffffffffc0000001, NTSTATUS (STATUS_UNSUCCESSFUL)
SYMBOL_NAME: dxgkrnl!DxgCreateLiveDumpWithDriverBlob+154
FAILURE_BUCKET_ID: LKD_0x1B0_dxgkrnl!DxgCreateLiveDumpWithDriverBlob
Stack: PnpDeviceActionWorker → PiRestartDevice → PipProcessDevNodeTree → PipCallDriverAddDevice → dxgkrnl!DxgCreateLiveDumpWithDriverBlob

Interpretation: the PnP manager forces a full device restart (PiRestartDevice) after failed TDR recovery, calls AddDevice on the driver, and the driver returns STATUS_UNSUCCESSFUL — the card fails to reinitialize after a software-triggered reset. This matches the Linux-side finding (amdgpu resume of IP block failed / GPU Recovery Failed) independently, on a completely different driver stack and OS.

WHEA-LOGGER checked for all incident time windows — zero platform/CPU/memory/PCIe-controller hardware errors logged, ruling out the motherboard/CPU/RAM as the source.

TROUBLESHOOTING ALREADY ATTEMPTED (none resolved the idle issue):

  • Linux: amdgpu.gfxoff=0 pcie_aspm=off amdgpu.runpm=0
  • Windows: PCIe ASPM disabled, MPO disabled, ULPS-related tweaks, TdrDelay/TdrDdiDelay increased to 8
  • Both physical BIOS profiles tested (OC and Silent)
  • Nearly the entire driver range from 24.3.1 through 26.7.1 tested (reviewed every release notes Fixed/Known Issues section in that range — no documented fix for this signature found)
  • DisplayPort and HDMI both tested
  • FreeSync/VRR disabled

Given this reproduces across two OSes, two BIOS profiles, and nearly every driver version spanning 2.5 years, with WHEA ruling out the platform, I don’t believe this is fixable from the user side. Has anyone seen this exact bucket ID (LKD_0x1B0_dxgkrnl!DxgCreateLiveDumpWithDriverBlob) resolved, or is this something the driver team would want the raw dump files for? I have the .dmp files saved and I’m happy to attach them.

Update: I’ve uploaded the raw watchdog dump files (.dmp) to Google Drive for analysis. You can access the folder here:

Your Linux log has a better root cause in it than the Windows bucket you’re leading with. “SMU: No response” is the first thing that fails, and everything after it, gfxoff not disabling, sdma timing out, the reset failing, is downstream of the power management processor being wedged.
On the Windows side, 0x1B0 add-device-failed is just the card not coming back afterwards. Re-titling around the SMU timeout instead of the dxgkrnl bucket will get you far better hits, and it’s the part a driver engineer actually wants. Worth filing at https://gitlab.freedesktop.org/drm/amd , that’s where the amdgpu devs read, and a cross-OS repro is exactly what gets picked up there.

Thank you very much! Hakob_Arzumanyan

1 Like