Frequent DX12 Device Removed (0x887A0005) and NaN Movement Errors

Description:

Since the latest patches, I’ve been getting very frequent DirectX 12 device removals, often preceded by errors related to NaN values in monster movement.
At first, I had generic [D3D12] Device Removed. Reason: 0x887A0005 crashes, which I mitigated with the following adjustments:

Added TdrDelay=10 to registry

Disabled Hardware-accelerated GPU Scheduling (HAGS)

Set renderer_skip_validation_layers=true and renderer_async_compute_enabled=false in production_Config.ini

Lowered texture filtering and disabled dynamic culling

Those steps reduced random GPU device timeouts, but the problem simply shifted: now I get more frequent NaN-related crashes or Failed to CreateResource errors right after a Device Removed.

System specs:

GPU: RTX 2080 Ti (11 GB)

Driver: NVIDIA 581.57 (clean install)

CPU: Ryzen 9 3900X

Resolution: 4K

OS: Windows 11 x64

Example log excerpts:
Spoiler

Tried to issue a MoveTo with NaN speed! Object: Metadata/Monsters/Breach/Monsters/FingerDemon/FingerDemon@80
[D3D12] Device Removed. Reason: 0x887a0005
[D3D12] Failed to CreateResource (error: 0x887a0005)


These usually happen during Breach or Abyss encounters, but I’ve also seen them in other dense areas.

What it looks like:

The game freezes instantly, then either blackscreens or returns to desktop.
After that, the renderer tries to restart ([RENDER] Restarting device), but fails to reinitialize resources.

Technical observation:

From the logs, it seems the engine is generating invalid (NaN) numerical values during AI or physics updates — possibly pathfinding or animation speed.
When those invalid values reach the DX12 renderer, the GPU command queue becomes invalid, triggering DXGI_ERROR_DEVICE_REMOVED (0x887A0005) or DXGI_ERROR_DEVICE_HUNG.

This isn’t a hardware or driver instability: it’s the engine sending impossible data to the GPU.
Switching to DX11 masks the issue (less strict validation), but doesn’t actually fix the source.

Request / feedback:

Could the dev team please investigate:

The root cause of NaN values in monster movement speed

The lack of error handling around invalid GPU command data

Resource allocation failures following a device loss (Failed to CreateResource)

The game is currently unplayable for more than a few maps at a time on DX12, even with stable hardware and conservative settings.

Thank you for looking into this — happy to provide full logs or reproduction details if needed.
Last bumped on Oct 19, 2025, 11:54:55 AM

Report Forum Post

Report Account:

Report Type

Additional Info