
NVIDIA doubles DLSS 5 speed and reduces VRAM usage by 20%
NVIDIA has announced an update to DLSS 5, which doubles the performance of the 3D-Guided Neural Rendering model and reduces VRAM consumption by 20% at 4K resolution.
What has changed in DLSS 5
The main improvement is related to the optimization of the neural network model responsible for realistic lighting and material detail. Unlike standard upscaling and frame generation technologies, DLSS 5 uses game engine data for additional AI-powered image processing.
NVIDIA engineers have redesigned the processing algorithms and the neural network structure. The model size has been reduced by 24% to approximately 118 million parameters, and VRAM consumption at 4K resolution has decreased by 20%.
As a result, the computational load on GPU has been halved compared to the September version. NVIDIA claims that image quality remains virtually unchanged.
Compared to the initial demonstration in March 2026, the speed of neural rendering has already increased tenfold. However, a twofold acceleration of the model itself does not mean a similar increase in FPS in games. The actual performance gain depends on the graphics card, graphics settings, and the load on other system components.
Which games will receive the updated DLSS 5
The optimized model will debut on October 16 in NBA 2K27. The update will be available to owners of GeForce RTX 50 series graphics cards, laptops on the RTX Spark platform, and users of the GeForce NOW cloud service. In the future, the new version will become standard for games with DLSS 5 support.
NVIDIA also confirmed the timeline for the technology’s arrival in other projects:
- Hogwarts Legacy – November 5. Developers will add DLSS 5 and update other DLSS technologies to their latest versions.
- Where Winds Meet – end of November. The game already supports DLSS Super Resolution and Multi Frame Generation, to which neural rendering will be added.
- RTX Remix – October 23. The toolkit for modifying classic games will receive an updated DLSS 5 and an accelerated frame generation mode.
NVIDIA paid special attention to Hogwarts Legacy. Developers at Avalanche Software tuned the neural network rendering to preserve the game’s artistic style. In the new demonstration, character appearances are noticeably closer to the original than in the March version of DLSS 5.
For RTX Remix, the company showcased the technology’s application in mods for Fallout 4 RTX and Mirror’s Edge RTX. Neural network rendering improves the display of reflections, shadows, lighting, and surfaces. Mod creators will be able to independently select objects and scene areas to which DLSS 5 is applied, as well as adjust the intensity of the changes.
Frame generation is 35% faster
At the same time, NVIDIA released a beta version of a new performance mode for DLSS Frame Generation, Multi Frame Generation, and Dynamic Multi Frame Generation.
The mode is called Preset C. In it, one of the two neural network models used for frame generation is replaced by a lighter algorithm. It processes optical flow for interface elements and ray tracing effects, reducing computational load.

On the GeForce RTX 5060, the new mode, according to NVIDIA, runs up to 35% faster than before. However, the acceleration comes with a slight degradation in image quality.
Unlike DLSS 5 itself, the new frame generation mode supports not only the GeForce RTX 50 series but also the RTX 40 series graphics cards, as well as RTX Spark laptops.
Preset C is already available in the beta version of the NVIDIA App. To activate it, you need to enable experimental updates in the application settings, then open the Graphics section, go to the DLSS Override – Model Presets parameter, and select Preset C for frame generation.
Official support for DLSS 5 neural rendering is currently limited to the GeForce RTX 50 series and the RTX Spark platform. NVIDIA previously announced plans to expand compatibility to the RTX 40 series after completing optimization for the current generation of graphics cards, but did not specify a timeline.











