AMD_WoW_0x141_Diagnostic
Issue: Intermittent DX12 GPU hang/TDR in World of Warcraft on Radeon RX 7900 XT.
SYSTEM
OS: Windows 11, build 10.0.26200.9168
GPU: Radeon RX 7900 XT
Current AMD driver: 32.0.31041.1004
Previous AMD driver: 32.0.31035.1003
Game: World of Warcraft, DirectX 12
PSU: 850W, 80+ Platinum
SUMMARY
The system intermittently experiences LiveKernelEvent 0x141 involving amdkmdag.sys while playing World of Warcraft in DX12.
The WoW client reports a Direct queue timeout / GPU Hung. Windows then records LiveKernelEvent 0x141 and performs TDR recovery. The GPU recovers successfully: D3D11 device creation succeeds shortly afterward and the adapter is re-enumerated.
The issue is NOT reliably reproducible.
Importantly, the same type of 0x141 failure has occurred on two AMD driver versions:
- 32.0.31035.1003
- 32.0.31041.1004
LATEST INCIDENT TIMELINE
03:02:46.139 World transfer / loading screen started
03:02:52.814 Load Map End
03:02:53.400 World Loaded 100% / loading screen disabled
03:03:10.217 WoW: Direct queue timeout → GPU Hung
03:03:10.265 WoW: device context lost / recovery started
03:03:13.269 Device destroy wait timeout
03:03:16.269 Device destroy wait timeout
03:03:20 Windows LiveKernelEvent 0x141 / WATCHDOG dump
03:03:23.278/.293 DeviceLost / D3D12CreateDevice failed with 0x887A0006
03:03:24.516 D3D11 device creation succeeded
03:03:26.614 GPU adapter re-enumerated
The GPU hang was detected approximately 17 seconds after the world/map transition completed. This is a temporal correlation only; the logs do not prove that the map transition itself caused the hang.
DX12 / WINDOWS RUNTIME CHECKS
No evidence was found that a missing or corrupted DX12 runtime caused the initial failure.
Verified:
- d3d12.dll present
- d3d12core.dll present
- dxgi.dll present
- d3dcompiler_47.dll present
- dxcore.dll present
- Relevant Microsoft binaries have valid Authenticode signatures.
- Direct D3D12CreateDevice(Feature Level 12_0) test returned S_OK.
- WoW successfully created a D3D12 device about 2.5 hours before the failure.
- No preceding D3D12/DXGI initialization or device-removed errors were found in the examined history.
The 0x887A0006 (DXGI_ERROR_DEVICE_HUNG) occurred AFTER the Direct queue GPU hang/TDR sequence and is therefore considered a consequence of the device-lost state, not evidence of a prior DX12 runtime initialization failure.
EVENT / HARDWARE EVIDENCE
In the examined period immediately before the hang:
- No WHEA-Logger events
- No PCIe/AER errors
- No Kernel-PnP errors
- No Windows display-driver recovery errors
- No dxgkrnl/dxgmms2/amdkmdag error preceding the Direct queue timeout
- No WoW fence, residency, allocation, shader, or device-removal error was logged before the hang.
GPU memory usage reported by WoW was approximately 7.2 GB / 19.2 GB (about 37%) before the incident; no evidence of VRAM exhaustion was found.
AMD DRIVER EVIDENCE
The latest 0x141 event was verified against the AMD kernel driver:
amdkmdag.sys base:
fffff805`a12a0000
WER P3:
fffff805`a1470fb0
Calculated offset:
+0x1d0fb0
The calculated address exactly matches the WER P3 address, confirming that the reported P3 address falls at offset +0x1d0fb0 within amdkmdag.sys for the latest incident.
Two earlier 0x141 events on driver 32.0.31035.1003 reported the same WER P3 address within the same boot session. The corresponding dump files were not accessible for direct lmvm verification, so their exact offset was not calculated.
WORKLOAD CORRELATION
The latest incident followed a normal WoW world/map transition:
- World transfer started at 03:02:46.139
- World loaded 100% at 03:02:53.400
- Direct queue GPU hang detected at 03:03:10.217
There were no logged errors between world loading completion and the GPU hang.
CURRENT ASSESSMENT
The evidence currently favors a GPU/driver/workload interaction at the Direct queue / GPU scheduler level rather than a DX12 runtime installation problem.
The logs identify the detection point (Direct queue timeout), but do not identify the exact GPU packet, command list, fence, or driver operation that stopped progressing.
A short ETW/WPR capture focused on DxgKrnl GPU scheduling around the next occurrence is planned to identify the last submitted/uncompleted GPU work and TDR/reset behavior.
REQUEST TO AMD
Please investigate the intermittent DX12 Direct queue GPU hang/TDR behavior in World of Warcraft on Radeon RX 7900 XT.
In particular, please check whether there is a driver issue involving the Direct queue / GPU scheduler / command submission or recovery path that can produce LiveKernelEvent 0x141.
The issue has occurred on two driver versions and is currently not reliably reproducible.
Available diagnostic evidence includes the LiveKernelEvent 0x141 / WATCHDOG dump and detailed WoW logs. An ETW GPU scheduler trace can be collected during a future occurrence if required.