4.6K ·GitHub
What is this?
DLSSG Native is a DLL mod that lets NVIDIA RTX 20 (SM75) and RTX 30 (SM86) series GPUs run DLSS Frame Generation in D3D12 games, even though NVIDIA doesn't officially support frame generation on those cards. It bundles its own C++ wrapper, PTX/Cubin kernels for both GPU generations, and the 310.1 inference model into a single DLL instead of using NVIDIA's original feature DLL.
DLSSG Native is a DLL mod that lets NVIDIA RTX 20 (SM75) and RTX 30 (SM86) series GPUs run DLSS Frame Generation in D3D12 games, even though NVIDIA doesn't officially support frame generation on those cards. It bundles its own C++ wrapper, PTX/Cubin kernels for both GPU generations, and the 310.1 inference model into a single DLL instead of using NVIDIA's original feature DLL.Is this a standalone app or something I drop into a game folder?It's a mod, not an app: you copy a proxy DLL (version.dll by default) and an INI config file (dlssgsm86.ini) into the folder containing the game's rendering executable. There's no installer or separate program to run.
It's a mod, not an app: you copy a proxy DLL (version.dll by default) and an INI config file (dlssgsm86.ini) into the folder containing the game's rendering executable. There's no installer or separate program to run.How do I install it into a game?Fully quit the game, find the folder containing the actual rendering EXE, and copy version.dll plus dlssgsm86.ini there (or use an alternate proxy name like winmm.dll, dinput8.dll, winhttp.dll or dxgi.dll if version.dll is taken). Pick Router=SM86 or SM75 in the INI for your GPU generation, then restart the game and enable DLSS Frame Generation in its settings.
Fully quit the game, find the folder containing the actual rendering EXE, and copy version.dll plus dlssgsm86.ini there (or use an alternate proxy name like winmm.dll, dinput8.dll, winhttp.dll or dxgi.dll if version.dll is taken). Pick Router=SM86 or SM75 in the INI for your GPU generation, then restart the game and enable DLSS Frame Generation in its settings.Do I need to install CUDA Toolkit or Python first?No. It only needs an NVIDIA driver exposing the NGX/NVAPI/CUDA interfaces (tested with driver 591.86); installing CUDA Toolkit or Python separately is not required.
No. It only needs an NVIDIA driver exposing the NGX/NVAPI/CUDA interfaces (tested with driver 591.86); installing CUDA Toolkit or Python separately is not required.Any catches I should know about before using it?It works by DLL proxying and LoadLibrary hooking, which heuristic antivirus engines can flag as suspicious even though the DLLs are self-signed and legitimate — SmartScreen may still warn. It also caps at 3 generated frames (2X/3X/4X) and does not support 6X, dynamic multipliers, Reflex Warp, or automatic Reflex matrix queries.
It works by DLL proxying and LoadLibrary hooking, which heuristic antivirus engines can flag as suspicious even though the DLLs are self-signed and legitimate — SmartScreen may still warn. It also caps at 3 generated frames (2X/3X/4X) and does not support 6X, dynamic multipliers, Reflex Warp, or automatic Reflex matrix queries.Is this limited to Windows and DirectX 12 games?Yes — it currently only works on Windows 10/11 x64 with D3D12 games; Vulkan support is planned for a future release but isn't available yet.
Yes — it currently only works on Windows 10/11 x64 with D3D12 games; Vulkan support is planned for a future release but isn't available yet.Which NVIDIA GPUs does it actually work with?The SM86 route targets the RTX 30 series and has been validated on a real RTX 3080 Ti; the SM75 route targets RTX 20 series/Turing cards but has only passed a forward-pass PTX test so far, with physical Turing/Cubin validation still pending.
The SM86 route targets the RTX 30 series and has been validated on a real RTX 3080 Ti; the SM75 route targets RTX 20 series/Turing cards but has only passed a forward-pass PTX test so far, with physical Turing/Cubin validation still pending.How much extra VRAM does frame generation use?On an RTX 3080 Ti, 0.2.4 adds roughly 320-340 MiB at 1080p, 490-520 MiB at 1440p, and 700-770 MiB at 4K depending on the 2X/3X/4X multiplier, on top of whatever VRAM the game itself needs.
On an RTX 3080 Ti, 0.2.4 adds roughly 320-340 MiB at 1080p, 490-520 MiB at 1440p, and 700-770 MiB at 4K depending on the 2X/3X/4X multiplier, on top of whatever VRAM the game itself needs.How much faster is 0.2.4 than the original 0.1.0 release?GPU frame-generation time dropped by about 26-35% across resolutions and multipliers on an RTX 3080 Ti; for example 4K at 4X fell from 7.115 ms to 4.804 ms per frame group.
GPU frame-generation time dropped by about 26-35% across resolutions and multipliers on an RTX 3080 Ti; for example 4K at 4X fell from 7.115 ms to 4.804 ms per frame group.Is Vulkan supported?Not yet — the current version only works with D3D12; Vulkan resource interop and synchronization are planned for the next release.
Not yet — the current version only works with D3D12; Vulkan resource interop and synchronization are planned for the next release.What are similar projects?DLSS5-Feeder, OptiScaler-DLSSNR-PreSR-Multipass, MFG-Unlocker, dlss-unlocked, 1-Click-DLSS5, DLSS-NR-on-AMD, p2-rtx, mcopt
DLSS5-Feeder, OptiScaler-DLSSNR-PreSR-Multipass, MFG-Unlocker, dlss-unlocked, 1-Click-DLSS5, DLSS-NR-on-AMD, p2-rtx, mcoptWhich topics is it in?dll-injection, frame-interpolation, frame-generation, performance-optimization, game-modding, modding, windows-10, windows-11, rendering, gpu-acceleration, game, graphics
dll-injection, frame-interpolation, frame-generation, performance-optimization, game-modding, modding, windows-10, windows-11, rendering, gpu-acceleration, game, graphics