[Bug] CreateFeature Access Violation (0xC0000005) on RTX 4060 (v1.10.0)

[deyonmagis-sudo]

Environment

• App Version: NeuralScreen 1.10.0
• OS: Windows 11 Pro 25H2 (Build 26200.9445)
• GPU: NVIDIA GeForce RTX 4060 (8GB VRAM, Ada Lovelace, arch 0x190)
• Driver: 32.0.16.1692 (GeForce 616.92 branch)
• Resolution: 1920x1080 (work_scale 0.65 -> 1248x702)

Logs

10:21:48.309  [arch] patch installed: 1 cards cached, architecture 0x190 (spoofed to the DLL's accepted value)
10:21:48.313  [pure] NGX calls go through C:\Users\ADMIN\Downloads\neuralscreen-v1.10.0-full\native\nvngx.dll_ns-forwarder.dll
10:21:48.475  [pure] DLSSNR Init_Ext (through the forwarder) -> 0x00000001 (Success)
10:21:48.549  [host] feature requirements query failed 0xBAD00012 - continuing with the create
10:21:48.709  [host] CreateFeature raised 0xC0000005 (caught; nothing submitted)
10:21:48.709  [video] NR feature unavailable (0x7FFFFFFF) - SAFE PASSTHROUGH

[perseval-BLR]

Thanks for the log - it is exactly what makes this diagnosable.

One line in it is a red herring: feature requirements query failed 0xBAD00012 is expected with the bundled runtime - its feature discovery data is newer than the build itself - and it is diagnostic only. The create below it is the real answer.

The create raising 0xC0000005 is the actual problem: the neural runtime faults inside its own feature creation on your 4060. Our worker catches the exception - the program does not crash, it falls back to passthrough - and keeps going without the effect. This is a real compatibility problem on our radar, not something you did wrong.

Two things would help a lot:

1. The full NeuralScreen.log from one clean run - the whole file. It opens with an environment header (driver, OS, HDR) that answers most of what we would otherwise have to ask.
2. One run per driver branch: an earlier GeForce branch and a Studio build. That tells us whether the create fault moves with the driver - your header shows 32.0.16.1692, and this stage is the one place our code cannot patch around.

One more variable you could try, if you have access to another runtime build: drop a newer NVIDIA-signed nvngx_dlssnr.dll into native\libraries\ next to the program. It overrides the bundled copy; the folder checks the signature before loading and names the reason in the log if it refuses.

The latest build is v1.11.0 - use it for the retest: https://github.com/perseval-BLR/DLSS5-NeuralScreen/releases/download/v1.11.0/neuralscreen-v1.11.0-full.zip

[perseval-BLR]

v1.12.0 is now available: https://github.com/perseval-BLR/NeuralScreen/releases/tag/v1.12.0

This release does not claim to make the underlying RTX 4060 runtime CreateFeature access violation disappear. It does make this path safer and easier to diagnose: the requirements query and CreateFeature are now reported as separate stages, and a terminal CreateFeature fault goes directly to safe passthrough without unsafe retries or resource teardown.

Please retest with v1.12.0 and attach the full new NeuralScreen.log. If it still reports CreateFeature raised, we will keep this open as a runtime/driver compatibility issue rather than treating the earlier 0xBAD00012 requirements result as the cause.

[perseval-BLR]

v1.13 does not claim to fix the NVIDIA runtime CreateFeature access violation. It adds an isolated three-frame 640×360 preflight before capture/presentation, fail-closed quarantine and a privacy-scrubbed diagnostic bundle. Please retest v1.13 on the RTX 4060 and attach the diagnostic bundle.

[deyonmagis-sudo]

Studio Driver (v596.36)

[env] NeuralScreen 1.12.0 | Windows 10.0 (build 26200) | Windows-11-10.0.26200-SP0
[env] GPU: RTX 4060 (40xx, arch 0x190)
[env] driver: 32.0.15.9636
[env] HDR: off
[env] Num Lock: on | lang: en | profile: Natural | work_scale: 0.65 | flow: fast
[main] monitor '\\\\.\\DISPLAY1' from config.json is not connected - using monitor 0
[main] monitor 0 is 1920x1080 (config: 3840x2160), taking the real resolution
[main] NeuralScreen - profile 'Natural', resolution 1920x1080, monitor 0
[main] NGX parameters: {'style': 1, 'auto_mask': 1, 'intensity': 1.0, 'local_tone': 0.5, 'local_structure': 1.0, 'skin_structure': -1.0}
[main] work_scale 0.65 (NGX resolution 1248x702)
[main] GPU: RTX 4060 · Ada (group 0x190, officially supported: no)
[main] pre-Blackwell GPU: warmup 120 -> 4 to avoid a false frame-0 watchdog timeout
18:42:45.769  [host] adapter 0: NVIDIA GeForce RTX 4060 vendor=0x10DE vram=7956MB luid=00000000:01820EA0
18:42:45.769  [host] adapter 1: AMD Radeon(TM) Graphics vendor=0x1002 vram=485MB luid=00000000:00016401
18:42:45.769  [host] adapter 2: Microsoft Basic Render Driver vendor=0x1414 vram=0MB luid=00000000:00016398
18:42:45.769  [host] adapter 0 selected by NS_GPU
18:42:45.769  [host] adapter 0 runs the network and the capture
18:42:45.944  [host] DRED settings unavailable (settings1=0x80004002, settings=0x80004002)
18:42:45.971  [pure] standalone D3D12 device ready; no swapchain or carrier modules
18:42:46.526  [host] NVSDK_NGX_D3D12_Init -> 0x00000001 (Success)
18:42:46.527  [arch] patch installed: 1 cards cached, architecture 0x190 (spoofed to the DLL's accepted value)
18:42:46.527  [pure] NGX calls leave the worker itself (the module path carries nvngx.dll)
18:42:46.926  [pure] DLSSNR Init_Ext (from the worker) -> 0x00000001 (Success)
18:42:46.926  [video] stream 1248x702 (full-res frames 1920x1080), LIVE (unbounded), warmup=4
18:42:46.976  [host] feature requirements query failed 0xBAD0000C - continuing with the create
18:42:47.146  [host] CreateFeature raised 0xC0000005 (caught; nothing submitted)
[main] shared memory agreed: 23.0 MB, the frame does not go through the pipe
18:42:47.146  [video] NR feature unavailable (0x7FFFFFFF) - SAFE PASSTHROUGH
[main] worker started (pid 14692), header sent (1248x702)
18:42:47.146  [video] SHMI: mapped 'NeuralScreen_11380_168be041', 23040000 bytes (motion at +8294400)
[main] capturing monitor 0: (1920, 1080)
[main] output window 1920x1080
[main] tray icon started
[main] taskbar window started
[main] hotkeys registered: Num1, Num2, Num6, Num4, Ctrl+Alt+Q, Num0, Num3, Num5, Num7 (Num1=toggle, Num2=settings, Num6=scale_up, Num4=scale_down, Ctrl+Alt+Q=quit, Num0=record, Num3=screenshot_menu, Num5=window_mode, Num7=framegen)
18:42:47.645  [present] hidden from screen capture
18:42:47.663  [present] overlay 1920x1080 ready (flip-discard, layered click-through, hidden from capture)
[main] presenting in the worker window: no frame comes back to Python
18:42:47.667  [present] hidden from screen capture
18:42:47.676  [dda] swizzle pipeline ready
18:42:47.676  [cap] adapter 0, the one the network runs on
18:42:47.695  [dda] output 0 selected by NS_OUTPUT (\\.\DISPLAY11)
18:42:47.695  [dda] output colour space 0, 8 bits per colour
18:42:47.696  [dda] SDR capture pinned to BGRA8 (stable across 10-bit scan-out)
18:42:47.696  [dda] desktop rotation: none
18:42:47.696  [dda] capture 1920x1080 active
18:42:47.696  [video] DDA1 OK (1920x1080)
18:42:47.721  [video] GRAY OK (320x180)
[main] gray channel 320x180: guides take luminance from the worker
[main] screen capture inside the worker (DDA1): no colour through the pipe
[main] motion field 320x180 - upscaled by the worker on the GPU
18:42:47.722  [video] OUTS OK (1920x1080)
[main] result pixels through shared memory (1920x1080, 8 MB)
18:42:47.723  [cap] capture format 87 (BGRA8)
18:42:47.723  [hdr] capture=SDR; neural processing=SDR proxy; export=SDR
18:42:47.723  [cap] shared texture 1920x1080 ready
18:42:47.727  [fg] UI: off, 2x
18:42:47.729  [present] window revealed on the first Present
[main] menu opened at startup
[main] NR ON | FPS 307.9 | frames 457 | work 1248x702 | scene 0.000
[main] NR ON | FPS 204.8 | frames 810 | work 1248x702 | scene 0.000
[perf] guides 0.2ms | send 0.0ms | recv 1.1ms | show 0.2ms
[main] NR ON | FPS  69.7 | frames 980 | work 1248x702 | scene 0.000
[main] NR ON | FPS 103.0 | frames 1298 | work 1248x702 | scene 0.000
[main] NR ON | FPS 269.6 | frames 1782 | work 1248x702 | scene 0.000
[perf] guides 0.1ms | send 0.0ms | recv 1.0ms | show 0.1ms
[main] NR ON | FPS 248.0 | frames 2173 | work 1248x702 | scene 0.000
[main] NR ON | FPS 193.0 | frames 2469 | work 1248x702 | scene 0.000

Game Ready Driver (v616.92)

[env] NeuralScreen 1.12.0 | Windows 10.0 (build 26200) | Windows-11-10.0.26200-SP0
[env] GPU: RTX 4060 (40xx, arch 0x190)
[env] driver: 32.0.16.1692
[env] HDR: off
[env] Num Lock: on | lang: en | profile: Natural | work_scale: 0.65 | flow: fast
[main] monitor 0 is 1920x1080 (config: 3840x2160), taking the real resolution
[main] NeuralScreen - profile 'Natural', resolution 1920x1080, monitor 0
[main] NGX parameters: {'style': 1, 'auto_mask': 1, 'intensity': 1.0, 'local_tone': 0.5, 'local_structure': 1.0, 'skin_structure': -1.0}
[main] work_scale 0.65 (NGX resolution 1248x702)
[main] GPU: RTX 4060 · Ada (group 0x190, officially supported: no)
[main] pre-Blackwell GPU: warmup 120 -> 4 to avoid a false frame-0 watchdog timeout
18:17:47.429  [host] adapter 0: NVIDIA GeForce RTX 4060 vendor=0x10DE vram=7956MB luid=00000000:00015123
18:17:47.429  [host] adapter 1: AMD Radeon(TM) Graphics vendor=0x1002 vram=485MB luid=00000000:00016401
18:17:47.429  [host] adapter 2: Microsoft Basic Render Driver vendor=0x1414 vram=0MB luid=00000000:00016398
18:17:47.429  [host] adapter 0 selected by NS_GPU
18:17:47.429  [host] adapter 0 runs the network and the capture
18:17:47.573  [host] DRED settings unavailable (settings1=0x80004002, settings=0x80004002)
18:17:47.589  [pure] standalone D3D12 device ready; no swapchain or carrier modules
18:17:48.206  [host] NVSDK_NGX_D3D12_Init -> 0x00000001 (Success)
18:17:48.206  [arch] patch installed: 1 cards cached, architecture 0x190 (spoofed to the DLL's accepted value)
18:17:48.206  [pure] NGX calls leave the worker itself (the module path carries nvngx.dll)
18:17:48.314  [pure] DLSSNR Init_Ext (from the worker) -> 0x00000001 (Success)
18:17:48.314  [video] stream 1248x702 (full-res frames 1920x1080), LIVE (unbounded), warmup=4
18:17:48.341  [host] feature requirements query failed 0xBAD00012 - continuing with the create
18:17:48.458  [host] CreateFeature raised 0xC0000005 (caught; nothing submitted)
[main] shared memory agreed: 23.0 MB, the frame does not go through the pipe
18:17:48.458  [video] NR feature unavailable (0x7FFFFFFF) - SAFE PASSTHROUGH
[main] worker started (pid 8136), header sent (1248x702)
18:17:48.458  [video] SHMI: mapped 'NeuralScreen_14224_4bfd3b09', 23040000 bytes (motion at +8294400)
[main] capturing monitor 0: (1920, 1080)
[main] output window 1920x1080
[main] tray icon started
[main] taskbar window started
[main] hotkeys registered: Num1, Num2, Num6, Num4, Ctrl+Alt+Q, Num0, Num3, Num5, Num7 (Num1=toggle, Num2=settings, Num6=scale_up, Num4=scale_down, Ctrl+Alt+Q=quit, Num0=record, Num3=screenshot_menu, Num5=window_mode, Num7=framegen)
18:17:48.822  [present] hidden from screen capture
18:17:48.835  [present] overlay 1920x1080 ready (flip-discard, layered click-through, hidden from capture)
[main] presenting in the worker window: no frame comes back to Python
18:17:48.840  [present] hidden from screen capture
18:17:48.845  [dda] swizzle pipeline ready
18:17:48.845  [cap] adapter 0, the one the network runs on
18:17:48.861  [dda] output 0 selected by NS_OUTPUT (\\.\DISPLAY1)
18:17:48.861  [dda] output colour space 0, 8 bits per colour
18:17:48.862  [dda] SDR capture pinned to BGRA8 (stable across 10-bit scan-out)
18:17:48.862  [dda] desktop rotation: none
18:17:48.862  [dda] capture 1920x1080 active
18:17:48.862  [video] DDA1 OK (1920x1080)
18:17:48.867  [video] GRAY OK (320x180)
[main] gray channel 320x180: guides take luminance from the worker
[main] screen capture inside the worker (DDA1): no colour through the pipe
[main] motion field 320x180 - upscaled by the worker on the GPU
18:17:48.867  [video] OUTS OK (1920x1080)
[main] result pixels through shared memory (1920x1080, 8 MB)
18:17:48.868  [cap] capture format 87 (BGRA8)
18:17:48.868  [hdr] capture=SDR; neural processing=SDR proxy; export=SDR
18:17:48.869  [cap] shared texture 1920x1080 ready
18:17:48.872  [fg] UI: off, 2x
18:17:48.873  [present] window revealed on the first Present
[main] menu opened at startup
[main] NR ON | FPS 106.4 | frames 279 | work 1248x702 | scene 0.000
[main] NR ON | FPS  94.7 | frames 596 | work 1248x702 | scene 0.000

[perseval-BLR]

v1.13.0 is out: NeuralScreen v1.13.0.

It changes the diagnostic side of the CreateFeature failure rather than claiming to fix it. Before any desktop capture, v1.13 runs an isolated preflight on three synthetic 640x360 frames. The native worker now reports the result of CreateFeature explicitly (CACK), and a failure, timeout, crash, TDR or device loss is a named, staged outcome with a timed quarantine - instead of a restart loop. Plain SAFE PASSTHROUGH pixels no longer count as evidence of support.

If you can, please retest on v1.13.0 and send:

1. NeuralScreen.log from the run (v1.13 writes the preflight stage and the reason);
2. a diagnostic package - Create diagnostic package on the Program tab. It contains the adapter, driver, stage, HRESULT/SEH/DRED markers and a scrubbed log tail, with usernames and paths removed.

That package is exactly what is needed to tell a runtime refusal apart from a driver or hardware-level AV, which are different fixes.

[deyonmagis-sudo]

Tested on v1.13.0 with the RTX 4060 (Game Ready Driver 616.92).

Preflight results

The synthetic 640x360 preflight check caught the CreateFeature Access Violation (0xC0000005) immediately during the create stage (failed: 0/0, expected 3).

[env] NeuralScreen 1.13.0 | Windows 10.0 (build 26200) | Windows-11-10.0.26200-SP0
[env] GPU: RTX 4060 (40xx, arch 0x190)
[env] driver: 32.0.16.1692
[env] HDR: off
[env] Num Lock: on | lang: en | profile: Natural | work_scale: 0.65 | flow: fast
[main] NeuralScreen - profile 'Natural', resolution 1920x1080, monitor 0
[main] NGX parameters: {'style': 1, 'auto_mask': 1, 'intensity': 1.0, 'local_tone': 0.5, 'local_structure': 1.0, 'skin_structure': -1.0}
[main] work_scale 0.65 (NGX resolution 1248x702)
22:28:53.343  [host] adapter 0: NVIDIA GeForce RTX 4060 vendor=0x10DE vram=7956MB luid=00000000:00014DF2
22:28:53.343  [host] adapter 1: AMD Radeon(TM) Graphics vendor=0x1002 vram=485MB luid=00000000:000160DA
22:28:53.343  [host] adapter 2: Microsoft Basic Render Driver vendor=0x1414 vram=0MB luid=00000000:0001607B
22:28:53.343  [host] adapter 0 selected by NS_GPU
22:28:53.343  [host] adapter 0 runs the network and the capture
22:28:53.502  [host] DRED settings unavailable (settings1=0x80004002, settings=0x80004002)
22:28:53.518  [pure] standalone D3D12 device ready; no swapchain or carrier modules
22:28:54.181  [host] NVSDK_NGX_D3D12_Init -> 0x00000001 (Success)
22:28:54.181  [arch] patch installed: 1 cards cached, architecture 0x190 (spoofed to the DLL's accepted value)
22:28:54.181  [pure] NGX calls leave the worker itself (the module path carries nvngx.dll)
22:28:54.309  [pure] DLSSNR Init_Ext (from the worker) -> 0x00000001 (Success)
22:28:54.309  [video] stream 640x360, LIVE (unbounded), warmup=0
22:28:54.309  [host] feature requirements query failed 0xBAD00012 - continuing with the create
22:28:54.472  [host] CreateFeature raised 0xC0000005 (caught; nothing submitted)
[compat] failed: 0/0 (expected 3), stage=create, reason=error, cached=no
[main] compatibility preflight cancelled - capture was not opened

NeuralScreen-diagnostics-20260916-222854.zip

[perseval-BLR]

Thank you for the follow-up with the full log - this is exactly the outcome the preflight exists for.

What your log shows, in order: feature requirements query failed 0xBAD00012 (expected with the bundled runtime - the leaked build does not expose the requirements query, and the program continues past it), then CreateFeature raising 0xC0000005 inside the isolated preflight, then failed: 0/0 (expected 3), stage=create and compatibility preflight cancelled - capture was not opened.

So on your 4060 the failure is now contained: no capture, no presentation, no half-initialised worker, and the verdict is cached so the next launch does not repeat it. That is the honest ceiling of what a fix on our side can do here - the access violation happens inside NVIDIA's runtime while creating the feature, on this GPU/driver combination, and nothing in our code is on the stack when it does. I am not going to claim otherwise.

Two things I would like to check in your diagnostics bundle, unless you have already looked: whether the DRED markers landed in it (your log shows DRED settings unavailable, which would mean no GPU crash dump), and whether the cached verdict is the one being served on the next launch. If you can also try the Studio driver you mentioned earlier on v1.13.1 - https://github.com/perseval-BLR/NeuralScreen/releases/tag/v1.13.1 - that would tell us whether this is driver-specific or a property of the card/runtime pair.