NVIDIA’s Deep Learning Super Sampling (DLSS) has become a staple for owners of recent graphics cards and the Nintendo Switch 2. The proprietary upscaling technology aims to enhance gaming performance by rendering scenes at a lower internal resolution and then using a machine learning algorithm to upscale the result to the monitor’s native resolution.
When integrated effectively into a game, DLSS allows players to crank up graphical settings beyond what their hardware would typically support at native resolution while maintaining playable frame rates. Alternatively, users can keep settings lower and push for higher frame rates, which is particularly beneficial for those using high-refresh-rate displays.
The Mechanics Behind DLSS
While the basic concept involves rendering fewer pixels and upscaling the image, understanding the specific modes helps users optimize their experience. The target resolution should almost always match the monitor’s native output. For instance, a 1440p monitor should have its display resolution set to 1440p.
DLSS operates through several machine learning models trained to interpret motion vectors, depth, and brightness data. Although NVIDIA has significantly improved these models over time, upscaling inherently involves a trade-off in image fidelity. The extent of this trade-off depends on the chosen DLSS mode:
- Quality: The GPU renders at 66.7% of the target resolution. On a 4K display, this means starting with a 1440p image, or 960p on a 1440p monitor. This mode delivers the best image quality with only a moderate performance boost.
- Balanced: Rendering occurs at 58% of the target resolution. At 4K, the internal resolution is 1253p, and 835p for 1440p displays. This offers a middle ground between visual clarity and higher frame rates.
- Performance: The GPU renders at 50% of the target resolution, equating to 720p on 1440p monitors and 1080p on 4K screens.
- Ultra Performance: Designed primarily for 8K monitors, this mode renders at just 33.3% of the output resolution. Using it on lower-resolution displays typically results in noticeable blurriness.
For 1440p users, Quality is generally the recommended starting point, with Balanced as a fallback for improved performance. 4K users often find Balanced to be the sweet spot, dropping to Performance if higher framerates are required.
A Suite of Related Technologies
DLSS branding now covers a family of distinct technologies that can be used in combination:
The original DLSS launched in 2019 alongside the RTX 20 series, initially limited to titles like Battlefield V and Metro Exodus due to the need for game-specific training. DLSS 2, released in 2020, removed this constraint, paving the way for widespread adoption. By early 2025, over 1,000 games and applications supported the technology.
In early 2025, NVIDIA introduced DLSS 4, which utilizes a new transformer-based machine learning architecture to improve image quality, reduce ghosting, and enhance stability. However, on older RTX GPUs, DLSS 4 may incur higher overhead, leading to worse performance compared to previous versions. To address this, NVIDIA provides presets allowing users to select older, more efficient models.
DLAA (Deep Learning Anti-Aliasing) functions as the opposite of upscaling. Instead of boosting frames, it takes a native resolution image and stabilizes it by reducing shimmer and flickering, serving as an enhanced temporal anti-aliasing solution. It can be enabled via the NVIDIA app even in games that do not natively support it.
Frame Generation, originally part of DLSS 3, allows RTX 40 series GPUs and newer to create interpolated frames, doubling the framerate when the GPU can consistently render 60 FPS or more. DLSS 4 expanded this with multi-frame generation for RTX 50 series GPUs, generating up to six interpolated frames for every one traditionally rendered frame.
Ray reconstruction, introduced with DLSS 3.5, is a denoising algorithm for ray-traced content. It samples light rays and generates high-quality pixels between them, primarily aiming to improve the coherence and appearance of ray and path tracing effects.
What is DLSS 5?
DLSS 5 represents a significant departure from traditional upscaling. Rather than targeting framerate improvements, it applies an AI filter to interpret color and motion vectors, adding photorealistic lighting and materials anchored to 3D content. Early demonstrations showed the technology enhancing character textures and surface details, though some critics noted an overly smoothed appearance.
Importantly, DLSS 5 does not improve performance; the computational overhead actually lowers frame rates. As of October 2026, the technology is officially available in only one title, NBA 2K27. While widespread developer adoption remains uncertain, modders have already begun porting the technology to various other games.
Leave a Reply