What is this?
d4r runs NVIDIA's official DLSS Super Resolution library inside Windows games on AMD Radeon GPUs under Linux and Proton, translating the DLSS network through ZLUDA (CUDA on ROCm/HIP) and replacing the heaviest kernels with hand-written RDNA3/RDNA4 code. It supports the DLSS 3 CNN model and the DLSS 4/4.5 transformer models; DLSS 5 is not supported.
d4r runs NVIDIA's official DLSS Super Resolution library inside Windows games on AMD Radeon GPUs under Linux and Proton, translating the DLSS network through ZLUDA (CUDA on ROCm/HIP) and replacing the heaviest kernels with hand-written RDNA3/RDNA4 code. It supports the DLSS 3 CNN model and the DLSS 4/4.5 transformer models; DLSS 5 is not supported.How do I install it for a game?Extract the release zip into the folder containing the game's .exe, then in Steam select GE-Proton 11 as the compatibility tool and add the launch options 'PROTONFORCENVAPI=1 DXVKNVAPIGPUARCH=AD100 %command%'. ROCm doesn't need to be installed separately, since its runtime is bundled in the zip.
Extract the release zip into the folder containing the game's .exe, then in Steam select GE-Proton 11 as the compatibility tool and add the launch options 'PROTONFORCENVAPI=1 DXVKNVAPIGPUARCH=AD100 %command%'. ROCm doesn't need to be installed separately, since its runtime is bundled in the zip.Do I need NVIDIA's own DLSS files to use this?Yes — you need NVIDIA's nvngxdlss.dll (310.7) and a matching nvngx.dll, since those files and the kernels built from NVIDIA's code aren't covered by d4r's own license. Building from source instead also needs ROCm with HIP, a GE-Proton build with OptiScaler integration, Rust, git-lfs, meson/ninja and MinGW/clang-cl toolchains.
Yes — you need NVIDIA's nvngxdlss.dll (310.7) and a matching nvngx.dll, since those files and the kernels built from NVIDIA's code aren't covered by d4r's own license. Building from source instead also needs ROCm with HIP, a GE-Proton build with OptiScaler integration, Rust, git-lfs, meson/ninja and MinGW/clang-cl toolchains.Is this ready for everyday gaming?No — the author describes it as a proof of concept, tested on only one GPU (an RX 7700 XT) in a handful of games, and it depends on unreleased patches to ZLUDA and vkd3d-proton. RDNA4 behavior is unverified on real hardware, and some games show visible artifacting with native upscaling.
No — the author describes it as a proof of concept, tested on only one GPU (an RX 7700 XT) in a handful of games, and it depends on unreleased patches to ZLUDA and vkd3d-proton. RDNA4 behavior is unverified on real hardware, and some games show visible artifacting with native upscaling.Will this work on my Radeon GPU?Only RDNA3 cards (RX 7700/7800/7900 series, gfx1100-1103) and RDNA4 cards (RX 9060/9070 series, gfx1200/1201) are targeted by the native kernels; RDNA2 and older GPUs are unsupported, and only the RX 7700 XT has actually been tested on real hardware so far.
Only RDNA3 cards (RX 7700/7800/7900 series, gfx1100-1103) and RDNA4 cards (RX 9060/9070 series, gfx1200/1201) are targeted by the native kernels; RDNA2 and older GPUs are unsupported, and only the RX 7700 XT has actually been tested on real hardware so far.How much does d4r actually speed things up?On an RX 7700 XT in SILENT HILL Townfall, DLSS Quality runs at roughly 67-72 fps versus 49.1 fps for native rendering, though it still trails AMD's own FSR 4 at the same settings.
On an RX 7700 XT in SILENT HILL Townfall, DLSS Quality runs at roughly 67-72 fps versus 49.1 fps for native rendering, though it still trails AMD's own FSR 4 at the same settings.Is this an official NVIDIA or AMD project?No, the README explicitly states d4r is not affiliated with NVIDIA or AMD.
No, the README explicitly states d4r is not affiliated with NVIDIA or AMD.How does the DLSS translation pipeline actually work?A D3D12 shim intercepts the game's NGX calls that OptiScaler forwards, a Wine-side CUDA bridge relays them to ZLUDA, ZLUDA translates NVIDIA's PTX code for the AMD GPU, and hand-written native kernels replace the heaviest DLSS network layers for speed.
A D3D12 shim intercepts the game's NGX calls that OptiScaler forwards, a Wine-side CUDA bridge relays them to ZLUDA, ZLUDA translates NVIDIA's PTX code for the AMD GPU, and hand-written native kernels replace the heaviest DLSS network layers for speed.What happens if native kernels aren't available for my GPU?The bridge falls back to ZLUDA's generic PTX translation, but without the native kernels, NVIDIA's translated DLSS 4 transformer layers produce invalid output that gets discarded, degrading image quality.
The bridge falls back to ZLUDA's generic PTX translation, but without the native kernels, NVIDIA's translated DLSS 4 transformer layers produce invalid output that gets discarded, degrading image quality.What are similar projects?CUDA-for-AMD-Windows, OptiScaler-DLSSNR-PreSR-Multipass, DLSS-NR-on-AMD, OptiScaler, LMCache, Hyperloom, DLSS5-Autopilot, cupy
CUDA-for-AMD-Windows, OptiScaler-DLSSNR-PreSR-Multipass, DLSS-NR-on-AMD, OptiScaler, LMCache, Hyperloom, DLSS5-Autopilot, cupyWhich topics is it in?proton, hip, dlss, wine, gpu-programming, amd-gpu, rocm, game-modding, amd, gpu-acceleration, graphics, gaming
proton, hip, dlss, wine, gpu-programming, amd-gpu, rocm, game-modding, amd, gpu-acceleration, graphics, gaming