Radeon GPU users no longer have to wait for AMD's response to NVIDIA's DLSS 5 Neural Rendering, as modders have delivered significant performance gains and simpler installation. Thanks to modder Daniel Blanco's unofficial project DLSS-NR-on-AMD, Radeon users can now run DLSS-NR with significantly higher frame rates and a much simpler, one-click installation.
There are actually two major runtime implementations at play here: Kien's dlss5-on-amd-9070xt-porting and danielblnc's DLSS-NR-on-AMD. Both are ground-up implementations that use custom HIP kernels to run the DLSS 5 neural network on Radeon hardware rather than translating NVIDIA GPU code at runtime. Daniel's approach remains closed-source while Kien's alternative is open-source.
Popular Now: MAXSUN removes the PCIe x16 slot from a Z890 Mini-ITX motherboard and routes the GPU over MCIO instead
According to danielblnc, the Alpha 0.3.3 version achieved a 23% speed boost over previous builds. Alpha 0.4.0 then increased performance by around 42%, and Alpha 0.4.1 added another 8%, bringing the total to roughly 89% more performance than v0.3.2 across those three releases. That is a massive improvement in just 2 weeks.
Kien's runtime also improved with AMDNR version 0.3.4. On an AMD Radeon RX 7800 XT at 1440p (FSR Quality), the update cut the neural network pass from 73 ms down to 52 ms, yielding a 29% drop in render time. RDNA 4 GPUs also see performance improvements, thanks to native support for FP8/E4M3 data types. There is also experimental support for handheld APUs like the Z1 Extreme / Z2 / Radeon 780M and newer 890M/880M-class hardware.
These performance boosts are further paired with the release of AMDNR Launcher, which handles installation and updates in just a few clicks, making it easier to use and install. It lets users choose between Daniel Blanco's runtime and Kien/lmxxf's HIP runtime while keeping the broader OptiScaler feature set available.
Previously, experimenting with this stuff involved copying OptiScaler builds, runtimes, weights, multiple DLLs, and configuration files into individual game directories.
At the end of the day, it is an independent attempt and has its caveats. Reimplementing DLSS 5 with custom HIP kernels is no magic fix. The neural workload remains expensive on Radeon hardware, even as developers increasingly exploit RDNA 4's native matrix and FP8 capabilities, and motion can still expose visual artifacts. The mod's reliance on DLL injection also restricts it to single-player titles.
Frequently Asked Questions
Open a question for an answer from TweakTown's coverage of this news, or ask your own below.
How do Daniel Blanco’s DLSS-NR-on-AMD and Kien’s AMDNR runtimes differ in performance, compatibility, and source availability?
Which single-player games can run DLSS 5 Neural Rendering on Radeon GPUs through OptiScaler?
How do users switch between Daniel Blanco’s runtime and Kien/lmxxf’s HIP runtime in AMDNR Launcher?
How does AMDNR performance vary across Radeon GPU models at 1440p?
Have a question that isn't listed here? Ask below, and TweakBot will answer it.
Ultimately, these mods show that community-driven engineering can bridge the feature gap long before hardware vendors do. Until AMD rolls out an official reply to NVIDIA's DLSS 5 Neural Rendering, tools like AMDNR Launcher offer Radeon owners a preview of where the industry may be headed.
