Deep Learning Super Sampling (DLSS) is a proprietary image reconstruction and upscaling technology developed by Nvidia for real-time graphics rendering, primarily in video games. It leverages Deep learning and Neural network models to reconstruct high-resolution images from lower-resolution inputs, aiming to improve frame rates while maintaining visual fidelity comparable to native rendering. The technology is integrated into Nvidia's GeForce RTX graphics cards and is a key feature of its RTX lineup, competing with similar upscaling solutions from other hardware vendors.
DLSS operates by rendering a game at a lower internal resolution and then using a trained neural network to upscale the image to the display's native resolution. The network is trained on high-quality, high-resolution frames to learn how to infer missing details, such as textures and edges, that are lost during downsampling. This process reduces the computational load on the GPU, allowing for higher frame rates or enabling ray tracing effects that would otherwise be too demanding. The technology has evolved through multiple versions, with each iteration improving image quality, temporal stability, and performance.
History and Development
Nvidia first announced DLSS in 2018 alongside its Turing architecture and the GeForce RTX 20 series. The initial version, DLSS 1.0, required per-game training and used a fixed neural network, which limited its adoption due to the need for game-specific optimization. In 2020, Nvidia introduced DLSS 2.0 with the Ampere architecture, which replaced the per-game approach with a general-purpose network that used temporal feedback from previous frames, significantly improving quality and ease of integration. This version became widely adopted across many titles.
In 2023, Nvidia released DLSS 3, which added frame generation technology. This feature uses an Artificial intelligence model to generate entirely new frames between rendered ones, effectively doubling or tripling the frame rate. DLSS 3 also introduced Nvidia Reflex, a latency reduction technology, to mitigate the input lag that can result from frame generation. The latest iteration, DLSS 4, announced in 2025, further refined the upscaling model with a transformer-based architecture, replacing the previous convolutional neural network, and introduced multi-frame generation for even higher performance.
Technical Approach
DLSS relies on a Deep learning model that is trained on a large dataset of high-resolution images and their corresponding low-resolution counterparts. The training process involves rendering scenes at various resolutions and using the high-resolution versions as ground truth. The network learns to predict the missing high-frequency details, such as texture patterns and fine edges, while also handling temporal information from motion vectors to reduce ghosting and artifacts.
The core of DLSS is a convolutional neural network, though DLSS 4 shifted to a Transformer (architecture)-based architecture, which is more adept at capturing long-range dependencies in the image. The model runs on dedicated tensor cores within Nvidia GPUs, which are specialized hardware for matrix operations, enabling real-time inference. The upscaling process is combined with temporal anti-aliasing, where the network uses information from previous frames to stabilize the image and improve quality over time.
Frame generation, introduced in DLSS 3, uses a separate neural network to interpolate between two consecutive rendered frames. This network estimates optical flow and generates a new frame that is inserted between them, increasing the perceived frame rate. The generated frames are not rendered by the GPU but are synthesized, which can lead to artifacts in fast-moving scenes if not handled carefully. Nvidia mitigates this by using motion vectors and a dedicated optical flow accelerator on its RTX 40 and 50 series GPUs.
Hardware and Software Integration
DLSS is exclusively available on Nvidia's GeForce RTX GPUs, starting with the RTX 20 series. The technology requires tensor cores, which are present in all RTX cards, and is supported through the Nvidia driver and the DLSS SDK. Game developers integrate DLSS into their engines using the SDK, which provides the upscaling and frame generation functions. Nvidia also offers DLSS as part of its broader RTX technology suite, which includes ray tracing and other AI-based features.
The software stack includes a plugin for popular game engines like Unreal Engine and Unity, making integration straightforward. DLSS is also used in professional applications, such as 3D rendering and video editing, where it can accelerate previews and final output. Nvidia regularly updates the DLSS model through driver updates, allowing for improvements without requiring game patches.
Comparison with Alternatives
DLSS competes with other upscaling technologies, notably AMD's FidelityFX Super Resolution (FSR) and Intel's Xe Super Sampling (XeSS). AMD's FSR is an open-source solution that works on a wider range of hardware, including older GPUs and even integrated graphics, but it does not use dedicated AI hardware, relying instead on spatial and temporal algorithms. Intel's XeSS uses a similar AI-based approach to DLSS but is designed for its Arc GPUs, though it also supports other vendors' hardware.
DLSS is often praised for its image quality and performance, particularly in its latest versions, but it is limited to Nvidia hardware. In contrast, FSR is more accessible but generally produces lower-quality results, especially at lower resolutions. The choice between these technologies often depends on the user's GPU brand and their preference for proprietary versus open solutions. As of 2025, DLSS remains the most widely adopted AI upscaling technology in the gaming industry.
Impact and Reception
DLSS has had a significant impact on the gaming industry by enabling high-fidelity graphics and ray tracing on consumer hardware. It has been credited with making ray tracing viable in real-time applications, as the performance boost from upscaling compensates for the computational cost of ray tracing. The technology has also influenced the development of other AI-based graphics techniques, such as frame generation and neural rendering.
Critics have noted that DLSS can introduce artifacts, particularly in early versions, and that frame generation can cause visual inconsistencies in fast-paced games. However, subsequent updates have largely addressed these issues, and DLSS is now considered a standard feature in many AAA titles. The technology has also been adopted in virtual reality and cloud gaming, where it helps maintain high frame rates on less powerful hardware. As of 2025, DLSS is a key differentiator for Nvidia's GPUs and a major factor in its market dominance in the high-end graphics segment.