Nvidia’s DLSS, or Deep Learning Super Sampling, is among the most controversial topics in PC gaming today: some love it, some hate it, and others are just confused. The battle often boils down to old-fashioned flame wars between fans of AMD and Nvidia, which can make the truth about DLSS even harder to understand.
Yet the reality of DLSS is simple once the hype and nonsense is tossed. DLSS is an excellent feature that can substantially improve performance, though it’s only useful if the PC games you enjoy support it.
- What is DLSS?
- Beyond Super Sampling: DLSS 3 and DLSS 3.5
- Why Does DLSS Matter for Gaming?
- DLSS vs. AMD FSR vs. Intel XeSS
- Conclusion
What is DLSS?
Nvidia DLSS is an acronym for Deep Learning Super Sampling. The acronym is out of date, however, because DLSS now includes features well beyond super sampling. DLSS has evolved into a suite of neural rendering techniques that use deep learning to improve image quality and performance.
The most basic implementation of DLSS, often referred to today as DLSS 2, examines frames and estimates what they should look like at a higher resolution (which is why Nvidia originally called the technique “supersampling”). It improves game performance by rendering games at a resolution below a display’s native resolution, then using AI to upscale the result.
Cyberpunk 2077 is an ideal example. If you select 4K resolution and choose DLSS Quality mode, the game will render at 1440p resolution. DLSS upscales the result to 4K. 1440p resolution requires less GPU grunt to render, so the game’s framerate improves.
However, it’s important to understand that DLSS uses neural rendering to upscale the image. This is a major change from older techniques, like checkerboard rendering. Because of this, DLSS can add detail that’s not visible when the game is rendered at native resolution with DLSS turned off, and it preserves details lost with other upscaling methods. It’s an impressive trick.
Beyond Super Sampling: DLSS 3 and DLSS 3.5
Supersampling isn’t the only trick up Nvidia’s sleeve. New versions of DLSS bring new features: DLSS 3 adds Frame Generation and DLSS 3.5 improves ray tracing.
DLSS 3 Frame Generation uses existing frames and deep learning to estimate what a new frame should look like. When this feature is on, DLSS 3 can add a new frame between each native frame.
It’s not unlike popular text-to-image apps, such as Dall-E 2 and Stable Diffusion. Though these apps primarily work with text input, they can also be used to upscale, expand, or alter existing images. That’s effectively what DLSS 3 does. It uses natively rendered frames as the basis for new AI-generated frames.
DLSS 3.5 (which is announced, but not yet released) will use deep learning to improve ray tracing performance. Modern ray tracing, though impressive, doesn’t cast enough rays to fill every pixel of a display—that would be too demanding. To get around this, games use “denoising” techniques to estimate the missing information.
It works, but the results are imperfect, which can lead to a grainy or chunky look. DLSS 3.5 tackles this problem with deep learning, which can learn what a scene should look like and better estimate the result. DLSS 3.5 is not released yet, so the scale of the improvement remains to be seen, but Nvidia’s preview indicates a major improvement in the clarity of ray-traced elements, such as reflections.
Why Does DLSS Matter for Gaming?
Understanding how DLSS works can be confusing, but the results are easy to understand. DLSS significantly improves performance and, in some cases, image quality.
IGN’s review of the RTX 4090 Founders Edition recorded an average frame rate of 42 FPS in Cyberpunk 2077 at 4K resolution and the Ultra Ray Tracing preset turned on. That’s playable, but far short of the 60 FPS most PC gamers prefer.
Turn DLSS on, however, and the story changes. DLSS 2 boosted performance to 84 FPS. That’s already excellent, but DLSS 3 put the game into overdrive, accelerating the average frame rate to an incredible 136 FPS. That’s a 300 percent improvement over native rendering. Who wouldn’t want that?
There are some caveats to this. DLSS 3 can experience latency issues with V-Sync turned on, or when used on entry-level hardware that can’t natively render the game beyond 30 frames per second. Still, DLSS 3 is impressive technology and well worth using in most games.
DLSS can also improve image quality. DLSS 2 supersampling can provide minor image quality benefits in some scenarios because, as mentioned earlier, it’s possible for Nvidia’s deep learning to introduce details not normally visible. This most often occurs along fine, high-contrast edges, like power lines or a chain-link fence. It’s not a sure-fire win, however, and exact results vary from game to game.
DLSS 3.5 is where image quality could really take a leap. Nvidia’s sample images for DLSS 3.5 show obvious improvements in the clarity of ray tracing. Light appears to scatter more naturally and reflections that appear mirror-like are much closer to how they would “really” look. This feature is not yet released, though, and it presumably will take some time for it to appear in a large number of games.
The key takeaway is this: DLSS can drastically improve a game’s performance and, in some situations, can slightly improve image quality. Given these benefits, DLSS should be used as frequently as possible.
DLSS vs. AMD FSR vs. Intel XeSS
Nvidia’s DLSS has competition. AMD’s alternative is FideltyFX Super Resolution (FSR), while Intel’s answer is Xe Super Sampling (XeSS).
DLSS holds a significant edge over its rivals due to its unique ability to create new frames. In contrast, Intel XeSS employs machine learning to enhance images from a lower render resolution to match your screen’s native resolution (much like Nvidia’s DLSS 2), and AMD’s FSR 2.0 utilizes a temporal upscaling method. AMD FSR 3.0, which is slated for release later in 2023, will add the ability to generate frames, though the method AMD will use to achieve this hasn’t been revealed.
In general, Nvidia’s DLSS has an edge in quality and performance. It boosts frame rates more aggressively than Intel XeSS and AMD FSR, and enhances image quality in ways they can’t. Direct comparisons of DLSS, FSR, and XeSS consistently give DLSS the edge.
But it’s not all good news.
Hardware and game support are an obvious problem for DLSS. It’s only supported on recent Nvidia graphics cards, and DLSS 3 requires the newest Nvidia RTX 40-series hardware. Game support is limited, too. The number of PC games with support for DLSS 2 numbers in the hundreds, which sounds like a lot, but leaves thousands of games without DLSS support. DLSS 3 support is even less common.
Nvidia’s approach stands in contrast to its rivals. AMD’s FSR doesn’t require AMD hardware, so it can run on Nvidia or Intel graphics. FSR does require support from a game’s developer, however, so it’s not universally available.
Intel’s XeSS also works on AMD and Nvidia hardware, though it works best on Intel hardware. It requires game support, too, and the list of supported games is rather short.
Conclusion
Nvidia’s DLSS can, in a best-case scenario, boost performance by up to 300 percent when compared to native rendering, and it does so without noticeably reducing image quality. On the contrary, it can sometimes improve image quality—a boon that will become especially noticeable once Nvidia’s DLSS 3.5, which improves ray tracing, begins to roll out in fall of 2023.
DLSS is superior to AMD FSR and Intel XeSS. Neither of Nvidia’s rivals can match its image quality, and they provide a less significant performance boost. This may change once AMD FSR 3.0 arrives in late 2023, however.
However, you need recent Nvidia hardware to use DLSS, and the game you’re playing must support it. That’s a sizable catch. DLSS isn’t useful if it’s not available in your favorite PC games.
That aside, DLSS is an excellent technology for PC gamers, and those who own Nvidia hardware should use it as frequently as possible. It can noticeably boost performance with little to no downside.
Matthew S. Smith is a hardware and tech freelance writer with more than 15 years of industry experience.
All content and images belong to their respected owners. This article is aggregated for informational purposes only with full credit to the source.
