D5 Lite for Rhino 1.2.0.0017 — D3D12 device removed on RTX 5090

molozero · September 23, 2026, 10:51pm

D5 Lite crashes roughly 20 seconds after launch with DXGI_ERROR_DEVICE_REMOVED (0x887A0005), reproducibly, more than ten times in a row. Turning off Hardware-Accelerated GPU Scheduling makes that crash disappear completely and the scene finally loads — the same Windows setting D5’s own help article tells users to turn on for DLSS Frame Generation.

Environment

Component	Version
D5 Lite for Rhino	1.2.0.0017
D5 Launcher	1.2.0.0110
D5 Render (full, installed alongside)	2.11.1.0623
Rhino	8.32.26160.13001
Windows	11 Pro, build 26200
GPU	GeForce RTX 5090, 32 GB
Driver	616.92 Game Ready (32.0.16.1692), WDDM 3.2

The machine is a Windows VM with the RTX 5090 passed through (QEMU/KVM, vfio-pci). That usually draws an “it’s the VM” answer, so up front: the full D5 Render 2.11 runs fine on this same machine, the host kernel log shows no PCIe errors and no device resets during any of the crashes, and long NVENC encodes run without incident. Details under Ruled out.

The crash

Every run has the same shape: D5 Lite opens, Rhino pushes the scene, it renders a few seconds at 20-30 fps, then one frame takes about nine seconds and the D3D12 device is gone. From D5Lite_2026-09-23-13-47-13.0.log:

13:47:23.38  [CookTask] resource RhinoStudio8.exr cook finished
13:47:24.01  camera_transition low fps: 1.74
             ...about 9 seconds with nothing logged at all...
13:47:33.33  camera_transition low fps: 0.11
13:47:33.33  [D3D12] Create Texture Resorce Failed With HRESULT -2005270523!
                With Name: HDRISkyBox   Size: 2048x2048   Format: 116
13:47:33.33  [D3D12] Format Support For this Format: RenderTarget 1 Read 1 Write 1
13:47:33.33  [D3D12] Create Texture With Fallback Driver API Failed! Please Update Your Driver
13:47:33.65  Crashed! Reason: Unhandled

-2005270523 is 0x887A0005, DXGI_ERROR_DEVICE_REMOVED. The app exits with code 10 and your crash reporter writes a report each time.

The failing texture is not consistent, which suggests it is simply the first allocation attempted after the device is already gone:

Run	Thread	Texture	Size
13:19	ModelViewer-VRAMIOService	unnamed	2048x1024
13:47	RenderThread	HDRISkyBox	2048x2048
13:58	ModelViewer-VRAMIOService	unnamed	2048x1024

In the Windows System log each crash is preceded by two or three nvlddmkm Error events, ID 153, about nine seconds apart. There is never an Event 4101 (“display driver stopped responding and has successfully recovered”), so Windows is not doing a TDR recovery here - the device is simply lost.

One crash report, if it helps you find it on your side: crash-report-8074103909904957275-2026-09-23-11-19-15Z.json, host_id 0ffdf150b3bc40cf97aaa7f64dc9da2d.

What turns it on and off: GPU scheduling

Turning off Hardware-Accelerated GPU Scheduling removes the device-removal crash completely. Settings > System > Display > Graphics > Change default graphics settings, or HwSchMode in HKLM\SYSTEM\CurrentControlSet\Control\GraphicsDrivers (1 = off, 2 = on). Reboot required.

With HAGS off:

no Create Texture Resorce Failed and no DEVICE_REMOVED anywhere in the log
the scene loads to completion for the first time: LoadProgress: event=scene_update phase=viewport status=ready degraded=false, sky 100%, material total=2 applied=2 unresolved=0
the session lasts about 90 seconds instead of 20

The GPU stall itself does not go away - nvlddmkm 153 and a low fps: 0.08 frame are still there. The difference is that the application now survives it instead of losing the D3D12 device.

This seems worth flagging on your side: your own help article, “How to enable Hardware-Accelerated GPU Scheduling”, tells users to turn this setting on, since DLSS Frame Generation requires it. On this machine that is precisely the configuration in which D5 Lite cannot run at all.

A second, separate crash (visible once the first one is gone)

With HAGS off the device removal is gone, but D5 Lite still dies about 90 seconds in, with a different signature: Crashed! Reason: SegFault, exit code 6 instead of 10. It follows 300 ms after a burst of window focus transitions:

14:07:41.5  [HostWindowPin] Foreground changed: kind=LiteMember
14:07:42.9  [HostWindowPin] Foreground changed: kind=HostMain
14:07:43.5  [HostWindowPin] Foreground changed: kind=External
            [HostWindowPin] Native band changed: old=Topmost, new=Normal, phase=Armed
14:07:43.6  [HostWindowPin] Foreground changed: kind=LiteMember
            [HostWindowPin] Native band changed: old=Normal, new=Topmost, phase=Elevated
14:07:43.9  Log.FATAL: Crashed! Reason: SegFault

There is no D3D12 error anywhere in that window, so this looks like a separate bug in HostWindowPinController (host_window_pin_controller_win.cpp), triggered by moving focus between the Lite window, the Rhino host window and a third application while Lite is pinned topmost.

Crash report: crash-report-346001432545811679-2026-09-23-12-07-44Z.json.

Ruled out

Each of these was tested rather than assumed:

Suspect	Test	Result
Stale GPU state from passthrough	Full host-level VM stop and start, GPU confirmed in D3hot before restart, Windows boot type 0x0	Crash identical
TDR	`TdrLevel` was 0 (detection disabled); set back to the Windows default 3, `TdrDelay` left at 10	No effect, and Event 4101 never appears
The sky HDRI	Removed the Rhino environment, so Lite fell back to its own `sky.hdr` (`Removed storage texture for Lite HDRI: RhinoStudio8.exr`)	Crash identical, on a different texture
VRAM	Failing textures are 2-32 MB on a 32 GB card, 600 MB used at idle	Not a capacity problem
Missing `bundles/dxil-shaders.bundle`	Logged as ERROR at every start, but compute pipelines are created successfully immediately after; reinstalling through the Launcher does not produce the file	Appears harmless
Host and PCIe layer	Host kernel log during every crash	No `vfio-pci` reset, no AER, no PCIe error
GPU health	`nvidia-smi` immediately after a crash	`CM_PROB_NONE`, 50 C, no HW thermal or power slowdown (0 us)
Lite vs full D5 Render	Same machine, driver, Rhino and day	Full D5 Render 2.11 ran repeatedly and exited cleanly; only Lite crashes

Happy to send the full D5 Lite logs, the dump files, or to run a build with extra logging if that helps.