Official Unity research: Vulkan follow-up, 2026-09-16
The project uses Unity 6000.3.24f1, revision 4e7b9b5b6244, and HDRP 17.3.0. Unity's release notes date this patch to 10 September 2026. The official release API, queried on 16 September, lists no newer 6.3 LTS patch. The current project is therefore not using an obsolete LTS patch.
HDRP 17.3 explicitly supports Windows with Vulkan. Unity's Vulkan manual says async compute and ray tracing are not implemented for Windows Vulkan. Compatibility alone does not establish stability of this project.
The release API without the LTS filter reports 6000.6.1f1, revision 7efac9f6c10e, released 2026-09-16T14:52:02.514Z. Its releaseNotes URL points to these official release notes. Both JSON and Markdown were retrieved directly with PowerShell because the newly published normal webpage was not yet retrievable through web browsing. The only listed graphics crash fix, UUM-150590, concerns per-material constant buffers on OpenGL ES. No documented matching Windows Vulkan/HDRP crash fix was found there. The prior full-game 6000.6.0f1 test already failed, while its minimal scene passed; see the local evidence summary.
| Official defect | Applicability and limits |
|---|---|
| UUM-101692: HDRP Vulkan standalone freeze | Multi-monitor/hybrid-output issue, commonly within ten seconds, already fixed in the 6.0/6.1/6.2 branches. Explicitly not reproduced on D3D11/12. Does not establish the cause of the project's cross-API failures. |
| UUM-87044: DrawBufferBatchMode crash | Windows/HDRP and D3D11/12/Vulkan; 6.3 status Open, fixed in 6.4.0b6/6.5.0a1. No matching local native stack established. Correction, 17 September: the historical 6.6 run used older content and cannot rule out this fix helping the current project; a content-matched upgrade comparison is still absent. |
| UUM-104763: constant-buffer scratch reuse | Open in 6.3, fixed in 6.6.0b1. Documented as D3D12 flicker, explicitly not reproduced on Vulkan/D3D11. Buffer-related vocabulary is not evidence that it explains the historical CBV violations or current resets. |
| UUM-151686: shadow-casting BatchRendererGroup crash | Open and reproduced in 6.3.24/6.6.1. Relevant only if the project's active path uses the required BatchRendererGroup and matches the native stack; neither condition is established by this research. |
No confirmed matching defect or promised upgrade cure was found. The local follow-up review establishes that disabling the SRP batcher alone has already failed, including 20260916-172258-graphics-stability with an effective post-initialization enabled=False pipelineReady=True log and a fresh NVIDIA TDR. Direct rendering alone also failed in 20260916-165916-graphics-stability. Neither switch is a demonstrated fix; repeating either alone adds little. A new combined comparison can test the simpler submission path, but any outcome applies to that combined configuration and cannot identify which component is responsible. A reported crash near a lap boundary does not by itself prove the lap transition caused it.
Passive recorder scope
-rampageCrashRecorder opts into a separate flight-recorder.jsonl. With -rampageReport, that folder must be unique per launch; an existing recording is never overwritten. Without it the recorder uses persistentDataPath/Diagnostics/<UTC-PID>/. It observes manual or automated play without steering it, samples at most once per second, records bounded Unity error messages, and flushes each record to disk. The recorder stops after 3,600 samples or 32 MiB and records the limit. A render callback is a CPU observation, not proof that the GPU completed/presented the frame. clean-exit means Unity invoked OnApplicationQuit, not a certification that the full game is stable. Native process termination can prevent the last sample or exit marker; errors absent from Unity's log callbacks require the ordinary player log and OS/dump evidence.