OptiScaler-MFG dev_8

OptiScaler-MFG dev_8

Pre-release. CI builds Windows x64 only; see Validation below for the author's GPU smoke test.

RTX 30 / 20 runtime correction

• RTX 20/30 MFG now uses the hash-pinned sdli1995/dlssg_for_sm86 0.3.5 version.dll as the primary game-side provider.
• The obsolete native dlssg_sm86.dll backend and its X5/X6 patch path were removed.
• OptiScaler no longer captures the game's nvngx_dlssg.dll load when the sdli1995 proxy is active.
• The release package places sdli1995 version.dll and dlssg_sm86.ini beside the game executable. Use OptiScaler.dll as winmm.dll, dxgi.dll, or another supported OptiScaler proxy name.

NVIDIA DLSS MFG unlock, native (RTX 40 / Ada)

• Blackwell (sm_120) kernels retargeted to Ada (sm_89), giving the game's own DLSS-G native multi-frame generation up to 6X.
• Transactional DLSSG 310.7/310.9 gates with rollback, temporal midpoint PTX correction, boundary artifact mitigation (modes 0/1/2, default balanced) and a silhouette guard.

RTX 30 / 20 (Ampere / Turing)

• sdli1995 0.3.5 is the primary RTX 20/30 runtime. The game-side plugin is capped safely when it only supports 4X.

Intel XeFG MFG

• Native 2X-6X unlock (five byte patches plus pacing per generated frame): [XeFG] UnlockMFG, MaxInterpolatedFrames, ExtraPacing.
• Menu options 2X-6X and the XeFG label.
• The game's DLSS-G is forced OFF when the output is XeFG/FSRFG, and swapchain recreation is fixed (RE Engine / Onimusha).

DLSS Neural Rendering (DLSS 5 / NR)

• Full migration to wilsjo's architecture: D3D12 multipass pre/post, native Vulkan, D3D11 bridge, finished-picture, GpuLifetime, CompatibilityRuntime, NgxDiagnostics and MenuModel.
• nvngx_dlssnr.dll runtime (ShortFuse RTX 20/30/40) bundled in the package by CI (release nr-runtime).
• Neural Rendering menu restored in the left column.

REFramework

• Native warning (toast plus menu header) when a RE Engine game lacks dinput8.dll; protected RE Engine titles need it to inject (tested on Onimusha: Way of the Sword).

Upstream ports

• Scaled-frame freeze detection with false frames (MFG pacing above 2X), typed DXGI exports without LOG_FUNC (Win11 crash), GpuTime _trigger timing, FSRFG fixes (swapchain/device), and older UE compatibility in FindFilePath.

Menu / unlocker

• Native DLSS MFG section shown only when FG Input and FG Output are None; FG selectors locked while the unlocker is active; correct save-and-restart prompt; Enable DLSS MFG Unlocker checkbox; noisy status messages removed.

Versioning

• Builds identify as OptiScaler-MFG dev_N (tag dev_8, asset OptiScaler-MFG_dev_8_<date>.7z).

Validation

• CI build: Windows x64. Author GPU smoke test: Onimusha: Way of the Sword (RE Engine) with REFramework - native RTX 40 MFG and XeFG OK, NR OK.

evairx/OptiScalerMFG dev-6

evairx/OptiScalerMFG dev-6

Normal release (not a pre-release). CI builds Windows x64 only; see Validation below for the author's GPU smoke test.

NVIDIA DLSS MFG unlock, native (RTX 40 / Ada)

• Blackwell (sm_120) kernels retargeted to Ada (sm_89), giving the game's own DLSS-G native multi-frame generation up to 6X.
• Transactional DLSSG 310.7/310.9 gates with rollback, temporal midpoint PTX correction, boundary artifact mitigation (modes 0/1/2, default balanced) and a silhouette guard.

RTX 30 / 20 (Ampere / Turing)

• Hash-pinned SM86 X5/X6 loader with the tested X4 fallback path.
• Experimental sdli1995 0.3.x runtime support (native 6X, [DLSSG] AmpereNative6XRuntime, default off).

Intel XeFG MFG

• Native 2X-6X unlock (five byte patches plus pacing per generated frame): [XeFG] UnlockMFG, MaxInterpolatedFrames, ExtraPacing.
• Menu options 2X-6X and the XeFG label.
• The game's DLSS-G is forced OFF when the output is XeFG/FSRFG, and swapchain recreation is fixed (RE Engine / Onimusha).

DLSS Neural Rendering (DLSS 5 / NR)

• Full migration to wilsjo's architecture: D3D12 multipass pre/post, native Vulkan, D3D11 bridge, finished-picture, GpuLifetime, CompatibilityRuntime, NgxDiagnostics and MenuModel.
• nvngx_dlssnr.dll runtime (ShortFuse RTX 20/30/40) bundled in the package by CI (release nr-runtime).
• Neural Rendering menu restored in the left column.

REFramework

• Native warning (toast plus menu header) when a RE Engine game lacks dinput8.dll; protected RE Engine titles need it to inject (tested on Onimusha: Way of the Sword).

Upstream ports

• Scaled-frame freeze detection with false frames (MFG pacing above 2X), typed DXGI exports without LOG_FUNC (Win11 crash), GpuTime _trigger timing, FSRFG fixes (swapchain/device), and older UE compatibility in FindFilePath.

Menu / unlocker

• Native DLSS MFG section shown only when FG Input and FG Output are None; FG selectors locked while the unlocker is active; correct save-and-restart prompt; Enable DLSS MFG Unlocker checkbox; noisy status messages removed.

Versioning

• Builds identify as evairx/OptiScalerMFG dev-N (tag dev-6, asset OptiScalerMFG_dev-6_<date>.7z).

Validation

• CI build: Windows x64. Author GPU smoke test: Onimusha: Way of the Sword (RE Engine) with REFramework - native RTX 40 MFG and XeFG OK, NR OK.

OptiScaler-MFG dev_7

OptiScaler-MFG dev_7

Pre-release. CI builds Windows x64 only; see Validation below for the author's GPU smoke test.

NVIDIA DLSS MFG unlock, native (RTX 40 / Ada)

• Blackwell (sm_120) kernels retargeted to Ada (sm_89), giving the game's own DLSS-G native multi-frame generation up to 6X.
• Transactional DLSSG 310.7/310.9 gates with rollback, temporal midpoint PTX correction, boundary artifact mitigation (modes 0/1/2, default balanced) and a silhouette guard.

RTX 30 / 20 (Ampere / Turing)

• Hash-pinned SM86 X5/X6 loader with the tested X4 fallback path.
• Built-in SM75/SM86 runtime is included and ready without a separate installation; the RTX 20/30 MFG unlocker remains opt-in to avoid unexpected game crashes. The game-side plugin is capped safely when it only supports 4X.
• Optional sdli1995 0.3.5 runtime support remains available through [DLSSG] AmpereNative6XRuntime.

Intel XeFG MFG

• Native 2X-6X unlock (five byte patches plus pacing per generated frame): [XeFG] UnlockMFG, MaxInterpolatedFrames, ExtraPacing.
• Menu options 2X-6X and the XeFG label.
• The game's DLSS-G is forced OFF when the output is XeFG/FSRFG, and swapchain recreation is fixed (RE Engine / Onimusha).

DLSS Neural Rendering (DLSS 5 / NR)

• Full migration to wilsjo's architecture: D3D12 multipass pre/post, native Vulkan, D3D11 bridge, finished-picture, GpuLifetime, CompatibilityRuntime, NgxDiagnostics and MenuModel.
• nvngx_dlssnr.dll runtime (ShortFuse RTX 20/30/40) bundled in the package by CI (release nr-runtime).
• Neural Rendering menu restored in the left column.

REFramework

• Native warning (toast plus menu header) when a RE Engine game lacks dinput8.dll; protected RE Engine titles need it to inject (tested on Onimusha: Way of the Sword).

Upstream ports

• Scaled-frame freeze detection with false frames (MFG pacing above 2X), typed DXGI exports without LOG_FUNC (Win11 crash), GpuTime _trigger timing, FSRFG fixes (swapchain/device), and older UE compatibility in FindFilePath.

Menu / unlocker

• Native DLSS MFG section shown only when FG Input and FG Output are None; FG selectors locked while the unlocker is active; correct save-and-restart prompt; Enable DLSS MFG Unlocker checkbox; noisy status messages removed.

Versioning

• Builds identify as OptiScaler-MFG dev_N (tag dev_7, asset OptiScaler-MFG_dev_7_<date>.7z).

Validation

• CI build: Windows x64. Author GPU smoke test: Onimusha: Way of the Sword (RE Engine) with REFramework - native RTX 40 MFG and XeFG OK, NR OK.

NR runtime (nvngx_dlssnr.dll)

DLSS Neural Rendering runtime for the CI of optiscaler_dev3+. ShortFuse compatibility build (RTX 20/30/40), SHA256 E67DEE209320CDAFE0E93E45675D7AA34323A53ACC57A72B2E40A181581C989A. Fetched automatically by just_build_no_signature.yml into external/dlss/ so the release package bundles it.

evairx/OptiScalerMFG dev-5

evairx/OptiScalerMFG dev-5

Normal release (not a pre-release). CI compiles Windows x64 only; nothing is runtime-verified on a GPU.

New in dev-5

• Fixed the save-and-restart prompt disappearing: the unlocker state was being latched a frame after the toggle by the Streamline hooks, which erased the red prompt immediately. It is now only latched where the patch actually lands (module load for Ada, sidecar load for Ampere), so the prompt stays until the game is restarted.
• Releases are normal releases from dev-5 on (tag dev-5, asset OptiScalerMFG_dev-5_<date>.7z).

New in dev-4

• Menu: turning Enable DLSS MFG Unlocker on or off now always shows the red "Save Settings and restart the game" prompt, and the status chatter (the green MFG Active: nvngx_dlssg.dll ... unlocked, the partial-match and the Ampere readiness lines) is gone. The unlocker state is latched once the patch lands, so the prompt only appears when a restart is really needed.
• Versioning: builds are named evairx/OptiScalerMFG dev-N (tag dev-N, asset OptiScalerMFG_dev-N_<date>.7z). No more 10.0.2 version prefix.

New in dev-3

• XeFG: the MFG selector now lists 2X, 3X, 4X, 5X and 6X as named options (values above 6X remain in a Custom slot), and the FG output is labelled just XeFG.
• Native NVIDIA MFG unlocker restored for the game's own DLSS-G. dev-2 only registered a native nvngx_dlssg.dll when the load was not redirected, so games whose DLSS-G is redirected to OptiScaler's modern package (Silent Hill 2 ships 3.7.0) never showed Enable MFG Unlocker and the unlock never applied. The module is now registered in both cases, the unlock is applied at module load like pre-1 (not only once the game enables DLSS-G), and the menu section appears whenever the game has its own DLSS-G and FG Input/FG Output are both None.
• Menu flow for the unlocker: renamed to Enable DLSS MFG Unlocker; while it is enabled the FG Input/FG Output selectors are locked out to avoid conflicts, and the native MFG section is hidden as soon as any FG input/output is selected.
• Ampere/Turing: the advertised multi-frame ceiling follows the SM86 X5/X6 patched-loader bound (up to 6X) instead of being clamped to 4X.
• NR (DLSS 5): nvngx_dlssnr.dll (ShortFuse RTX 20/30/40) is bundled in the release package, fetched by CI from the nr-runtime release or the NVNGX_DLSSNR_URL repository variable.
• Branding: builds identify as evairx/OptiScaler-MFG dev-3.

New in dev-2

• Native XeFG MFG unlock (2X to 6X) — no external ASI. Ports upstream's in-memory libxess_fg.dll patch set (five verified byte patches with rollback) plus per-generated-frame pacing, so Intel MFG no longer falls back to 2X on non-Intel adapters and 5X/6X stay selected. [XeFG] UnlockMFG, MaxInterpolatedFrames (default 5 = 6X) and ExtraPacing. V-Sync is required above 4X.
• DLSS Neural Rendering ("DLSS 5") — D3D12 stage 1. Native port of the NR pass from the GPL fork: runs over what the upscaler just wrote (native DLSS or OptiScaler's own upscalers), on the same command list, before frame generation consumes it. Adds the [DlssNr] config block and the Neural Rendering menu section. Vulkan/bridges and the scan/compare tooling are not in this stage.

Carried from dev-1

• Native NVIDIA MFG unlocker separated from autonomous FG: off by default, only offered for the game's own DLSS-G, mutually exclusive with OptiFG backends.
• Ada MFG: temporal midpoint correction + transactional patching with rollback, native-module-only targeting.
• Ampere/Turing: hash-pinned X5/X6 loader patcher with automatic fallback to the proven X4 path.
• Autonomous FG Output list now only exposes AMD FSR FG 2X and Intel XeSS 3 XeFG.

Nightly Builds

Changes in the Last 24 Hours

No changes...