NVIDIA's DLSS Multi Frame Generation (MFG) is officially exclusive to the RTX 50 'Blackwell' series, with simple Frame Generation limited to RTX 40 and above. However, modders have now managed to run NVIDIA's actual DLSS (Multi) Frame Generation code, rather than substitute technologies like AMD's FSR Frame Generation, on three older GeForce generations: Ada Lovelace (RTX 40), Ampere (RTX 30), and Turing (RTX 20).
The RTX 40 series has the most mature support, with several unlockers such as Dashdogy's RTX40MFG-Unlock, Mavismmg's MFGAdaUnlock-RenoDx, and matiasLombo's mfg-unlock offering fixed and experimental MFG modes for games that already support DLSS Frame Generation via NVIDIA's Streamline framework.
For RTX 30 GPUs, developer sdli1995 created dlssg_for_sm86, which implements a backend for Ampere's SM86 compute capability, embedding NVIDIA's DLSS MFG 310.1.X runtime alongside a custom backend. It has been tested on an RTX 3080 Ti and supports 2X, 3X, and 4X modes where available, though compatibility issues, missing options, crashes, and visual artifacts have been reported in some games.
The Turing-based RTX 20 series has now joined via Coldwood1026's dlssg_for_sm75, an experimental fork of the Ampere project adapted for SM75. It swaps GPU kernels and modifies architecture checks so the runtime reads Turing GPUs as Ada Lovelace. The mod has been verified on an RTX 2060 Max-Q running The Witcher 3's DirectX 12 next-gen update, where the in-game Frame Generation toggle worked and improved framerate was observed. Results are mixed and early: one Reddit user reported that 2X mode worked fine on an RTX 2060, but 3X showed ghosting and 4X did not hold up, with performance slightly below FSR Frame Generation on the same card. Both sources caution that broad compatibility should not be assumed and that sample sizes for these reports remain small.
All of these mods are described as experimental and could break with future game, driver, NVIDIA App, or DLSS runtime updates, since NVIDIA retains control over the closed-source NGX and Streamline DLSS MFG plugin; an earlier over-the-air update already temporarily disrupted Ada unlock methods before modders adapted.
Separately, modders have also been working on reducing the performance cost of DLSS 5 on RTX 50 GPUs. Version 0.7.1 of the OptiScaler-DLSSNR-PreSR-Multipass project introduced a hybrid FP8/NVFP4 version of DLSS 5's Neural Rendering model, aiming to take advantage of Blackwell's native NVFP4 Tensor Core acceleration. According to the project's release notes, this reduced the Neural Rendering pass time at 4K by only about 1-2% compared to the original FP8 version, a gain the developer called 'VERY minor,' and this does not directly translate to overall game performance improvement. A later version, 0.7.2, made the hybrid NVFP4 path the recommended option for Blackwell, showing a small FPS difference in a Baldur's Gate 3 test that the developer said has not been established as repeatable. NVIDIA has told Wccftech that DLSS 5 is already about five times faster than its original GTC 2026 demonstration, with further improvements and RTX 40 series support planned.
Correction — 2026-09-11: Clarified that the article analyzes developer-reported measurements rather than presenting the outlet’s own performance test.