NVIDIA‘s Deep Learning Super Sampling (DLSS) represents a powerful, proprietary upscaling technology designed to enhance game performance. When utilized correctly, it significantly boosts the graphical capabilities of modern gaming setups.
How DLSS Functions
DLSS works by fundamentally changing how a game’s image is rendered. Instead of drawing the game at the final, high monitor resolution, the GPU first renders the game at a reduced internal resolution, meaning it processes fewer pixels. Following this rendering, NVIDIA’s machine learning algorithm then reconstructs and increases the image quality to the monitor’s native resolution. This process is driven by an AI model trained by NVIDIA.
When implemented effectively within a game, DLSS allows players to achieve visual settings that would otherwise exceed the GPU’s normal capabilities at native resolution. Alternatively, users can run games at lower settings to maximize the framerate, which is especially beneficial for those utilizing high refresh rate monitors.
DLSS Modes and Resolution Scaling
While DLSS technology can seem magical, its operation involves a technical compromise. The machine learning models interpret complex data points, including motion vectors, depth information, and brightness details, when upscaling an image. Although NVIDIA has continually improved these models, some loss of visual fidelity is inherent in the upscaling process. The degree of loss depends on the specific DLSS mode selected, as this mode determines the initial, lower internal rendering resolution.
The primary modes and their respective scaling factors are as follows:
- Quality: The GPU renders the game at 66.7 percent of the target output resolution. For example, on a 4K monitor, the internal rendering starts at 1440p, while on a 1440p monitor, it starts at 960p. This mode provides the best image quality among the presets but yields only a moderate increase in framerate.
- Balanced: The GPU renders the game at 58 percent of the target output resolution. At 4K, this means a starting point of 1253p. For a 1440p monitor, the internal rendering starts at 835p. This mode aims to strike a middle ground between high image quality and a significant boost in framerate.
- Performance: The GPU renders the game at 50 percent of the target output resolution. This translates to a 720p internal resolution when used with a 1440p monitor, or a 1080p internal resolution when targeting 4K output.
- Ultra Performance: The GPU renders the game at 33.3 percent of the output resolution. This mode is generally recommended for use with 8K monitors, but using it with 1440p or 4K monitors may result in a noticeably blurry image.
As a general guide, for users with 1440p monitors, DLSS Quality is a suggested starting point. If higher frame rates are desired, users may drop down to Balanced mode, though further reductions risk significant blurriness. For 4K monitors, Balanced mode is a good initial choice, with Performance mode available if more frames are necessary.
Advanced DLSS Features and Technologies
The DLSS brand has expanded to encompass several related AI technologies. The initial version of DLSS appeared in 2019 with the release of NVIDIA’s RTX 20 series GPUs. Initially, only a small number of games supported it because NVIDIA had to train the algorithm specifically for each game. Since the introduction of DLSS 2 in 2020, this limitation has been removed, and as of the beginning of 2025, over 1,000 games and applications support DLSS.
A major technological advancement occurred at the start of 2025 with the release of DLSS 4. NVIDIA updated the algorithm to leverage a new transformer-based machine learning architecture, resulting in improved image quality, reduced ghosting effects, and better overall image stability. Due to a higher operational overhead, running DLSS 4 on older RTX GPUs may result in lower performance compared to previous DLSS versions; consequently, NVIDIA provides users with alternative DLSS “presets” to access older, more efficient models.
Other related features include:
- Deep Learning Anti-Aliasing (DLAA): Unlike DLSS, which boosts performance by sacrificing some quality, DLAA focuses on improving image stability. It takes a native output and reduces visual artifacts like shimmering or flickering, offering an enhancement over traditional temporal anti-aliasing (TAA).
- Frame Generation: Introduced with DLSS 3, this feature allows NVIDIA RTX 40 series GPUs and newer to synthesize an interpolated frame between two traditionally rendered frames, thereby increasing the game’s overall framerate. It performs best when the GPU can maintain a consistent 60 FPS or higher, and the user possesses a high-refresh rate monitor.
- Multi Frame Generation: With DLSS 4, NVIDIA introduced multi frame generation, enabling RTX 50 series GPUs to generate and interpolate a greater number of frames based on a single rendered frame. As of the release of DLSS 4.5 in March, the current maximum limit is six generated frames for every rendered frame.
- Ray Reconstruction: DLSS 3.5 introduced this denoising algorithm, designed to minimize visual artifacts associated with ray tracing. It improves the coherence and quality of scenes by sampling the light rays and generating high-quality pixels in the spaces between those rays.
DLSS 5: Advanced AI Application
DLSS 5 represents a significant departure from the upscaling methods described previously. Instead of improving resolution, DLSS 5 interprets a game’s color and motion vectors to apply what NVIDIA describes as “photoreal lighting and materials that are anchored to source 3D content and consistent from frame to frame.” Essentially, DLSS 5 functions as an advanced AI filter applied to the game.
While early demonstrations showcased an impressive visual effect, the technology does not boost the frame rate; rather, it lowers it due to the processing overhead required. As of October 2026, DLSS 5 is officially available in only one game, NBA 2K27. It remains uncertain whether developers will adopt DLSS 5 widely, though modders have begun integrating the technology into various titles.
