Skip to content

NR: restore 200% model raster and add D3D12 inspection hold - #2

Open
grim-susemi wants to merge 2 commits into
y4my4my4m:dlss-neural-renderingfrom
grim-susemi:restore-nr-200pct-holdframe
Open

grim-susemi wants to merge 2 commits into
y4my4my4m:dlss-neural-renderingfrom
grim-susemi:restore-nr-200pct-holdframe

Conversation

@grim-susemi

Copy link
Copy Markdown

Restores the 200% model raster cap and adds a transient D3D12 inspection Hold frame.

200pct (3 files): Model resolution slider and deferred commit go 100 to 200 in DlssNr_Menu.cpp; runtime caps 1.0f to 2.0f in DlssNr_Dx12.cpp and DlssNrFeature_Vk.cpp. No config, INI, forwarder, or shader-filtering change.

Hold frame (3 files, new design, no history to cherry-pick): Hold frame / Resume / Cancel hold control beside Capture 8 frames in DlssNr_Menu.cpp; InspectionHoldState request/query/release API in DlssNrFeature_Dx12.h; matched GPU snapshot with inspection-only resolve while held, history reset on resume, capture serialized in DlssNr_Dx12.cpp. Session-only, no config or INI change. D3D12 only; unavailable on native Vulkan and the proxy path.

Known caveat: above 100% the box-area upsample replicates source pixels with a bilinear resolve, so 200% costs four times the model pixels without adding game geometry samples. GPU runtime unverified. Release x64 build: 0 errors, OptiScaler.dll 25917440 bytes, SHA-256 E6C7C6ED6DC4BB4DC399E55FD7064CDD4987E1BC8FEB888F5BA299819479EE5D.

The Unlock MFG on RTX 40 checkbox and patch-status lines sat inside the
dlssgGameDMFGSupported block, so games/SL versions reporting no Dynamic
MFG hid the entire unlock UI. Move the unlock section out of that gate;
keep only the DMFG FPS-target slider gated. Patch logic untouched.
Model resolution slider and both runtime caps go back to 25-200%.
Hold frame is a new transient D3D12 inspection hold: the next
successful frame's proxy, model answer and original are snapshotted
together, Compare and Debug stay interactive, and the model does not
run again until Resume. Session-only, no config or INI change.

Caveat: above 100% the box-area upsample replicates source pixels
with a bilinear resolve, so 200% costs four times the model pixels
without adding game geometry samples. GPU runtime unverified.
SirenBrink pushed a commit to SirenBrink/OptiScaler_DLSSNR_Multipass_MFG_FFXIV that referenced this pull request Sep 10, 2026

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant