Nvidia DLSS: Pushing the Boundaries of AI-Powered Upscaling

Nvidia’s Deep Learning Super Sampling (DLSS) technology continues to evolve, demonstrating remarkable capabilities in upscaling images from extremely low resolutions. Recent experiments, notably by YouTuber 2kliksphilip, have showcased DLSS’s performance when reconstructing visuals from as low as 38×22 pixels to 4K resolution . While such low resolutions aren’t practical for gameplay, these tests illuminate the power of AI in image reconstruction and the advancements made since the introduction of DLSS 2.0 in 2020 .

The Extreme Upscaling Experiment

2kliksphilip’s experiment involved rendering scenes from Kingdom Come Deliverance 2 at just 1% of the standard internal resolution. This resulted in an initial image of approximately 38×22 pixels, representing roughly 0.01% of the pixels found in a 3840×2160 4K frame . At this scale, the game world appears as a collection of colored blocks, though user interface elements, often rendered separately at output resolution, remain relatively sharp.

The experiment highlights how DLSS functions by leveraging prior frames, motion vectors and per-frame buffers to reconstruct a detailed image from minimal input data. Rather than attempting to create a perfect image from a single, tiny frame, DLSS intelligently fills in the gaps, demonstrating its ability to infer detail and structure.

When Does DLSS Become Usable?

While upscaling from 1% resolution yields a visually poor result, the tests reveal a threshold where DLSS begins to produce recognizable and usable images. According to 2kliksphilip, a minimum internal resolution scale of around 20% is needed for games to become playable, depending on the title and the amount of motion . At a 4K output, this equates to an internal resolution of 768×432, representing 4% of the total pixel count.

This reduction in shading function offers significant performance gains, allowing gamers to achieve higher frame rates without sacrificing visual quality. The technology’s effectiveness stems from its utilize of AI and the dedicated Tensor Cores found in GeForce RTX GPUs .

The Evolution of DLSS

The advancements demonstrated in these upscaling experiments underscore the progress made with DLSS since the release of DLSS 2.0 in March 2020 . DLSS 2.0 marked a significant leap in AI rendering, offering improved performance and visual fidelity in supported games like MechWarrior 5: Mercenaries and Control.

The ongoing development of DLSS continues to push the boundaries of what’s possible in real-time image upscaling, offering gamers a pathway to higher resolutions and smoother frame rates.