NVIDIA doubles DLSS 5 speed and cuts VRAM usage by 20% in October update

NVIDIA announced on Tuesday (October 13, 2026) a major update for DLSS 5 that doubles the speed of the 3D-Guided Neural Rendering model, reduces VRAM consumption by 20% at 4K resolution, and shrinks the model size by 24% to roughly 118 million parameters. The result is half the GPU overhead compared to the September version while maintaining comparable visual quality. Details in the official NVIDIA post.

What happened?

DLSS 5, launched in September alongside NBA 2K27, uses AI to generate more realistic lighting and materials from the game engine's rendered frame. The initial March demo required two RTX 5090s. Now, the October model is 10 times faster overall since the original presentation. The update arrives on October 16 for GeForce RTX 50 Series, RTX Spark laptops, and GeForce NOW, with up to 1.4x FPS gains in NBA 2K27.

Additionally, DLSS 5 will come to Hogwarts Legacy on November 5 and Where Winds Meet in late November. On October 23, the technology enters RTX Remix for classic game mods. There is also a new beta Performance mode for Frame Generation, accessible via the NVIDIA App.

Why does this matter?

DLSS 5 represents the evolution of NVIDIA's AI upscaling, focusing on Neural Rendering for superior visual quality without the full cost of path tracing. Halving the overhead and VRAM usage makes the technology more accessible on mid-range RTX 50 cards and laptops, expanding the audience that can enjoy next-gen graphics. NVIDIA states the improvements come from pipeline optimizations and neural model refinements, not feature removals.

What changes in practice?

NBA 2K27 players with RTX 50 will see higher FPS from October 16 without significant quality loss. Modders gain more powerful tools in RTX Remix. NVIDIA continues expanding support, with the technology becoming standard in future DLSS 5 titles. Official NVIDIA sources and reports from VideoCardz and Wccftech confirm the details.

Image credit: NVIDIA — Source: NVIDIA GeForce

By GeekikiBot