What is NVIDIA DLSS?

Nvidia launched its own image enhancement and scaling technology in 2019 called DLSS (Deep Learning Super Sampling). It’s exclusive to Nvidia RTX graphics cards, due to the use of the Nvidia neural network. More than 300 Games support DLSS and it’s growing every month.

At CES 2026, Nvidia announced the latest version of the technology, DLSS 4.5, with some performance gains exclusive to the 5080, 5090, 5070 Ti, and 5070.

You can check out our 50-Series explainer page for more info.

By activating the DLSS you could expect a better image quality and higher FPS. The jump will depend on the NVIDIA GPU RTX you have, as availability of some function depend of the RTX generation

DLSS is an execution of TAAU (temporal anti-aliasing up-sampling), which uses data from previous images through sub-pixel variation to create fine detail and to reduce aliasing.

Since launch, Nvidia keeps updating the logic behind. Data considered for DLSS 2.0 includes: low-resolution raw input, motion vectors and exposure/brightness information.

programmed heuristic method. As a result, DLSS can generally resolve details better than other TAA and TAAU implementations, while suppressing most temporal artifacts. This is why it can sometimes produce a sharper image than rendering at higher, or even native, resolutions using traditional TAA. However, no temporal solution is perfect, and artifacts (ghosting in particular) are still visible in some scenarios when using DLSS.

Given that temporal artifacts occur in most art styles and environments in the same way, the neural network that powers DLSS would not need to be retrained for new games. Despite this, Nvidia frequently delivers new minor revisions of its DLSS for new titles through drivers update, suggesting that minor training optimizations are carried out as games are released.

DLSS Version 3.0 introduces motion interpolation, which allows the image generation algorithm to use two successive images to create a third one in between to smooth out movements. This also means that for every new image rendered, a new image is generated.

For the time being, however, it remains exclusive to RTX 4000 series graphics cards, due to the inclusion of a new-generation Optical Flow Accelerator.

DLSS 3 also incorporates NVIDIA Reflex, which synchronizes the GPU and CPU, ensuring optimum responsiveness and low system latency. Lower system latency makes game controls more responsive, When compared to native, DLSS 3 can reduce latency by up to 2X. (NVIDIA Relfex sub function works with GTX 900 Series or newer.)

The addition of ray reconstruction arrives with DLSS version 3.5, replacing several denoising algorithms with a single AI model

DLSS 4 is not exclusive to the 50-Series GPUs, but they will certainly provide the best utilisation of it. Team Green has continued to lean on AI, with it looking like artificial intelligence will provide the bulk of performance increases going forward as opposed to good old fashioned brute-force compute. The 5090 specifically, features 3352 AI Tensor cores, which is almost double that of the 5080. So if you're really keen to get the most out of DLSS 4, AI Tensor cores appear to be the new currency for that.

Announced at CES 2026, DLSS 4.5 brings new AI assisted improvements. As with DLSS 4 this works best with Nvidia 50-Series GPUs but can benefit anyone using a 20-Series or later. These features include Ray Reconstruction, Super Resolution and Multi Frame Generation.

Ray Reconstruction was first introduced with DLSS 3.5 and has seen a significant improvement since. The Second-Generation Transformer model means it no longer relies on game engine denoising solutions. Instead it reconstructs ray traced lighting using AI models to produce less blurred or smudged textures than normal denoising solutions.

Overall, Ray Reconstruction produces cleaner, more realistic ray-traced visuals by using AI which is trained offline using large datasets allowing more accurate predictions in real time, rather than using a hand crafted filtering technique.

DLSS 4.5 helps solve the challenges around Temporal Anti-Aliasing (TAA) and previous versions of Super Resolution. These methods often resulted in smudged or blurred textures and strange lighting anomalies. DLSS 4.5 Super Resolution by comparison uses a new, more powerful AI model. It helps remove shimmering and ghosting that could occur with older versions of DLSS. The AI model now has 5X the compute power of previous versions.

This should mean less ghosting, clearer visuals with 4K upscaling, and visuals that can meet, and sometimes exceed the native resolution of your display.

The original Multi Frame Generation introduced with DLSS 2 allowed for 2X sampling, meaning for every image rendered, a second was also rendered to improve image quality. With DLSS 4.5, Dynamic Multi Frame generation means up to 6X sampling, not only does this mean more frames are rendered. But importantly it can be processed 35% faster than before. All this means you might actually stand a chance at maxing out a 240hz gaming monitor without having to tank your settings or play at a super low resolution.

Version 1.0: Feature of the RTX 20 series with only two games supported, the first iteration of DLLS requires specific AI training for each game independently.

Version: 1.9 or 1.bis: this major update replaces the DLSS assignment to the Tensor Core with an assignment to the CUDA Cores Shader.

Version 2.0: An AI-accelerated version of TAAU, using Tensor Cores, and trained in a generic way for all games.

Version 3.0: Addition to the previous version of an optical flow image generation algorithm to double the refresh rate (only available on RTX 40 series).

Version 3.5: Added ray reconstruction by replacing the various algorithms with a single AI model trained on a dataset five times larger than version 3.0.

Version 4.0: Adds Multi Frame Generation (Exclusive to the 50-Series.)

Version 4.5: NVIDIA DLSS 4.5 Ray Reconstruction, second generation AI model, Dynamic Frame Generation and Super Resolution