OVERCLOCK.news
Live Feature
Feature gaming

DLSS 5 isn't an upscaler. Here's what NVIDIA says it actually does

NVIDIA's research page describes DLSS 5 as a one-step diffusion model that generates the final image instead of reconstructing it, and at SIGGRAPH it showed the dials developers get to rein it in.

By OVERCLOCK September 16, 2026 3 min read 5 sources
BlueskyXRedditMail

Every earlier version of DLSS has been, at heart, a very clever approximation. The game renders less than your screen needs, fewer pixels or fewer frames, and a neural network fills in what a more expensive render would have looked like. DLSS 5 changes the job. On 1 September NVIDIA's Applied Deep Learning Research group published a page setting out how it works, and its central claim is that DLSS 5 is "the first DLSS technology to generate the final displayed appearance", rather than reconstruct it.

NVIDIA's split-screen comparison of Grace Ashcroft from Resident Evil Requiem on a rainy street, labelled DLSS 5 Off on the left and DLSS 5 On on the right, where the lighting on her face and jacket looks richer.
NVIDIA's own before and after, from March. The right half is generated by DLSS 5 from the game's rendered frame on the left.Image: NVIDIA

Generate it from what? From the game's own frame, plus what the model has learned about how the real world looks. NVIDIA calls these learned appearance priors: knowledge picked up from real-world visual data about how light and materials behave, which no amount of extra rendering samples can efficiently supply. The game still decides what is in the scene and where it is. DLSS 5 changes how it looks, and NVIDIA's page stresses that it complements the game's renderer and its artists rather than replacing them.

The model is a diffusion model, from the same family as the image generators that turn text into pictures. Those usually start from random noise and remove it a little at a time, over many steps, each one a full pass through a large network. That is fine when you are happy to wait a few seconds for a picture and hopeless when a game running at 60 frames per second has under 17 milliseconds for everything. NVIDIA's version runs in a single step. It works directly on pixels, where many image generators work in a compressed intermediate form, and it is not working blind: for every frame it is given the rendered image, the engine's motion vectors, state carried over from previous frames, and the values the game's artists have set.

Two more words on that page do a lot of work: causal and deterministic. Causal means each frame is made only from the present and the past, never from frames that haven't happened yet, which a live game could not provide anyway. Deterministic means the same inputs always give the same output. An ordinary image generator adds randomness on purpose, which is why the same prompt gives a different picture every time; in a game that randomness would show up as shimmer from one frame to the next. NVIDIA says the model is also trained specifically for frame-to-frame stability, and runs inside a strict compute budget for every frame, at up to 4K.

How DLSS 5 makes a frame What it is given every frame Rendered frame the game's own image Motion vectors from the game engine Carried state memory of past frames Artist settings model, sliders, masks One-step pixel-space diffusion model causal, deterministic, trained for stability draws on appearance learned from real imagery Final displayed frame up to 4K, on GeForce RTX 50 Series Earlier DLSS: approximate a costlier render. DLSS 5: generate the final appearance.
Four inputs, one step, one frame. The model is conditioned on what the game rendered, so it enhances that frame rather than inventing a new one.Diagram: OVERCLOCK, from the inputs listed on NVIDIA ADLR's DLSS 5 page, 1 September 2026.
View as text
How DLSS 5 makes a frame
StageWhat it is
InputsThe rendered frame, engine motion vectors, state carried from earlier frames, and artistic-direction values
ModelA one-step pixel-space diffusion model, causal and deterministic, trained for frame-to-frame stability
OutputThe final displayed frame, at up to 4K, on GeForce RTX 50 Series GPUs

What NVIDIA spent SIGGRAPH on was control. At its keynote in Los Angeles on 20 July, NVIDIA listed three challenges for DLSS 5: preserving artistic intent, frame-by-frame streaming and real-time 4K, according to PCGuide, which reported the slides. Developers get three trained models, Model A, Model B and Model C, which interpret a scene differently and can be mixed across a game by scene, environment, object or character. Club386, which also covered the session, reports NVIDIA's reasoning: there is no single correct reference image for these scenes, so more than one result can make sense, and the developer has to choose.

On top of the models come two global sliders running from 0 to 1, Structure Intensity and Tone Intensity, which Club386 describes as setting how strongly DLSS 5 works across the complex and simple parts of a scene. There is also developer masking, with automatic detection of models and characters, so an artist can leave the effect off a particular character or object. Characters were the obvious concern: Grace Ashcroft from Resident Evil Requiem was a leading example in NVIDIA's March demonstration, and PCGuide notes that with masking a studio could simply leave its character models out.

What developers get to control, as shown at SIGGRAPH
Control What it does
Model A, B or CThree trained models that interpret a scene differently; mix them by scene, environment, object or character
Structure IntensityA global slider from 0 to 1
Tone IntensityA global slider from 0 to 1
Developer maskingTurn the effect on or off for individual objects or characters, with automatic detection of models and characters
A dial, not a switch. NVIDIA's answer to the question of who decides how a game looks: the studio, per scene and per object.Sources: PCGuide and Club386, 21 July 2026, reporting NVIDIA's SIGGRAPH 2026 keynote.

NVIDIA's research page says DLSS 5 runs on GeForce RTX 50 Series GPUs. NVIDIA said in March that it would arrive this fall, with support from publishers including Bethesda, CAPCOM, Tencent, Ubisoft and Warner Bros. Games, and in games including Starfield, Resident Evil Requiem, Assassin's Creed Shadows and Hogwarts Legacy. Studios plug it in through Streamline, the same framework they already use for DLSS and NVIDIA Reflex.

What has not been published is the cost. PCGuide reports that the March preview needed two RTX 5090s and that NVIDIA is optimising DLSS 5 to run on one GPU, but neither NVIDIA's research page nor the SIGGRAPH coverage gives a frame-time or memory cost, and we could find no independent measurement. That is the number to watch when it ships. Until then the fair summary is NVIDIA's own: a generative stage inside the rendering pipeline, working alongside the game's renderer, and a genuinely new kind of DLSS.

Sources

Every figure in this article traces to one of the following. Where a manufacturer and an independent test disagree, both are listed and the disagreement is stated in the text rather than resolved silently.

  1. NVIDIA ADLR DLSS 5: Generative Neural Rendering Published 1 September 2026: a one-step pixel-space diffusion model conditioned on the rendered frame, motion vectors, carried temporal state and artistic-direction values; causal, deterministic, trained for temporal stability, under a strict per-frame budget at up to 4K; the first DLSS to generate the final displayed appearance; runs on GeForce RTX 50 Series GPUs; complements rather than replaces rendering and artistic authorship.
  2. NVIDIA NVIDIA DLSS 5 Delivers AI-Powered Breakthrough in Visual Fidelity for Games 16 March 2026: arriving this fall; publisher and game support; integration through the Streamline framework used by DLSS and Reflex. Also the image in figure 1.
  3. NVIDIA At SIGGRAPH, NVIDIA Advances Graphics and Simulation With Agentic and Physical AI The NVIDIA keynote at SIGGRAPH 2026 in Los Angeles on 20 July, on neural rendering, world models and simulation.
  4. PCGuide Nvidia just showed off more DLSS 5 to explain how it preserves artistic intent, with controls for developers The three challenges on NVIDIA's slides; Models A, B and C; Structure Intensity and Tone Intensity sliders from 0 to 1; developer masking with automatic detection; Grace Ashcroft as a leading example in the March demo; the March preview needing two RTX 5090s and being optimised for one GPU.
  5. Club386 Nvidia unveils DLSS 5 game developer controls, stressing it respects artistic intent 21 July 2026: three models with varying training data and parameters, chosen per environment, object and character; NVIDIA's explanation that without a reference ground truth several outcomes can make sense; sliders setting strength across complex and simple elements.

This is original work by OVERCLOCK, not a summary of another outlet's reporting. Specifications are as published by the manufacturer unless a measured figure is named as such; vendor performance claims are labelled with their precision and are not comparable across different numeric formats. Prices and availability change, and are accurate as of publication.

More hardware