NVIDIA's DLSS: What is it and how to make the most of it

MicroCenterOfficial · June 2023

Written by Jon Martindale

Deep Learning Super Sampling, or DLSS, is NVIDIA’s flagship upscaling algorithm that can dramatically improve in-game performance. It’s the culmination of years of development in game optimization, dynamic settings adjustments, and artificial intelligence. When used effectively, it can more than double your frame rate without you having to lower a single setting. In short, DLSS can be amazing.

It’s not perfect, though. There are some image degradation issues, particularly on thin, perforated textures and objects, and the latest frame generation feature of DLSS 3 also has its own unique artifacts to contend with. You also need to have one of the latest generations of RTX graphics to make the most of it, although older generations of DLSS are supported by a wider range of GPUs.

But as more generations of RTX cards are released, and the onboard hardware that powers DLSS gets better, it’s becoming a much more widely adopted feature, and one that stands to benefit anyone gaming on an NVIDIA GPU. Today, hundreds of games support DLSS, making it possible for gamers with high-end GPUs to enjoy the smoothest of possible games, and those with older and more entry-level cards the option of playing newer games that their hardware just wouldn’t be capable of otherwise.

What is DLSS?

DLSS is an image enhancement and upscaling technology that uses machine learning and the Tensor cores on NVIDIA’s RTX graphics cards to dynamically raise the perceived resolution on your games, whilst actually rendering at a lower resolution. That allows the game to run at a higher frame rate without as great an impact on perceived image quality, as it would do if you actually lowered the resolution. Alternatively, gamers can enable other detail settings, such as ray tracing, without having such as a dramatic effect on their frame rate.

The DLSS technology is something that has to be implemented by game developers, so not all games support DLSS. It must also then be enabled in-game by gamers, who typically have the option of four different levels of DLSS. Quality mode looks the best, but has the least impact on frame rate by employing a scale factor of 1.5 times the actual rendered resolution. Balanced, Performance, and Ultra Performance modes render the game at lower and lower resolutions, closing the perceived visual gap with more aggressive upscaling. That leads to better performance the further down the scale you go, at the cost of image quality.

DLSS was first introduced with the RTX 2000 graphics cards in 2019 with a limited selection of games. Although early adoption among gamers and developers was slow, it quickly gathered apace, and with the launch of competing standards from AMD and Intel, and ongoing developments of DLSS, it can now be found in hundreds of games. These include early standouts like Metro Exodus and Control, as well as much-newer games, like Cyberpunk 2077, Plague Tale: Requiem, and Hogwarts Legacy.

The latest generation of deep learning super sampling is DLSS 3, and it provides a unique frame generation feature which further enhances game performance beyond that of DLSS 2 and other upscaling techniques. Where other generations of DLSS merely adjusted the resolution and applied effects to retain detail whilst reducing demand on the GPU, DLSS 3 uses AI to artificially create entire frames. That has its own drawbacks, since the AI is creating effectively new information, but it’s using a lot of in-game data to create that new frame and since it’s slotted between two real frames, it’s harder to notice. The overall effect is a big jump in performance, massively improving the fluidity and perceived frame rate of the gameplay experience.

How does DLSS work?

There are several components to the way DLSS fits into the render pipeline for a graphics card, and that’s changed significantly over the years as NVIDIA has released new, more advanced versions of its DLSS technology and the Tensor cores that accelerate it.

The first iteration of DLSS, known as DLSS 1.0, used the game’s last frame and motion vectors (the difference in viewpoint between the current and previous frame) to perform edge enhancement and spatial anti-aliasing. This takes the game at a lower rendered resolution, sharpens its edges, and then smooths out any harsh corners. This was then combined with data from a single frame rendered at a much higher resolution to create a final frame.

The results were impressive, but the loss in image quality could be significant, with fine details often lost or corrupted in the process.

DLSS 2.0 works a little differently, using temporal anti-aliasing upsampling and much more data from previous frames to help resolve finer details, so they aren’t lost in the upscaling process. It also leverages a new neural network to identify and fix temporal artifacts caused by the upscaling, allowing DLSS 2 to occasionally produce an image that is sharper and clearer than even rendering it at native resolution.

DLSS 2 also uses the Tensor cores more efficiently, leading to much reduced overhead and better overall performance, with a reduced impact on frame times. It also doesn’t have to be implemented on a game by game basis in conjunction with NVIDIA, so developers have a much easier time adding DLSS 2 support to their game than they did with the first generation. This led to an explosion on the use of DLSS among developers and eventually, gamers, with many more games adding support for it throughout and beyond this generation of DLSS.

DLSS 3 is the latest generation of DLSS and debuted alongside NVIDIA’s RTX 4000-series graphics cards. It is the most advanced version yet and makes use of the new optical flow accelerator in NVIDIA’s 4000-series GPUs to further improve performance in supporting games.

The big new feature with DLSS 3, though, is optical-flow frame generation. It takes two rendered frames from the rendering pipeline, and uses the power of artificial intelligence to generate a frame that sits between those two, making the transition between frames smoother, and enhancing the frame rate of the game.

This can have a dramatic impact on performance, leading to a more than doubling of frame rates over native rendering in some cases. That can be especially important in games with advanced ray tracing, like Cyberpunk 2077’s Overdrive Mode. In NVIDIA’s launch of its RTX 4000 graphics cards, it showed Cyberpunk running at just 22 frames per second on an RTX 4090 when running natively at 4K with ray tracing enabled, but DLSS turned off. Enabling DLSS 2 saw the game’s frame rate boost to over 60 FPS, but switching to DLSS 3 with frame generation saw the FPS skyrocket to over 90.

As the newest of the DLSS versions and its restrictions to RTX 40-series hardware, DLSS 3 use among developers and gamers is in the nascent stages of adoption. However, there are 10s of titles planning to add DLSS 3 support in the near future, so its usage is likely to increase dramatically throughout 2023 and beyond.

What graphics card do you need to use DLSS?

DLSS is a feature that requires the use of NVIDIA’s Tensor cores. Those are a specific hardware chip found only on NVIDIA’s RTX graphics cards, so you cannot use DLSS on NVIDIA GPUs that were released before the RTX 2000 series, or any others branded as “GTX” rather than “RTX.” That includes GTX 900 and GTX 1000-series GPUs, as well as the more recently released GTX 16-series.

That also means that DLSS does not work on AMD or Intel graphics cards, nor either companies onboard GPUs.

The list of NVIDIA graphics card generations that support DLSS is as follows:

RTX 2000 – E.g. RTX 2060, RTX 2070 Super, RTX 2080 Ti.
RTX 3000 – E.g. RTX 3050, RTX 3060 Ti, RTX 3080.
RTX 4000 – E.g. RTX 4070 Ti, RTX 4080, RTX 4090.

What graphics card do you need for DLSS 3?

If you’re excited to play one of the latest games using DLSS 3, then you’ll need one of the latest graphics cards. DLSS 3 is the first generation of DLSS since its debut to require a whole new generation of graphics card. While RTX 2000 and RTX 3000 graphics cards can use DLSS 2.0 and all of its iterations just fine, only RTX 4000 graphics cards can support DLSS 3 and its new frame generation feature.

The reason for this is because RTX 4000 cards are built with a new optical flow accelerator, which wasn’t incorporated in earlier RTX generations.

The list of graphics cards and mobile GPUs that support DLSS 3 at the time of writing, are:

NVIDIA RTX 4090
NVIDIA RTX 4080
NVIDIA RTX 4070 Ti
NVIDIA RTX 4070
NVIDIA RTX 4060 Ti
NVIDIA RTX 4050 (laptop only)

Any future RTX 40-series graphics cards and GPUs in desktops and laptops are also likely to support frame generation.

What games support DLSS 3?

The list of games that supports DLSS 3 is currently limited to a small selection of games, but it is growing all the time. At least as many games again have been announced with upcoming DLSS 3 support, so the list should be much more expansive by the end of the year.

Here are the games that support DLSS 3 at the time of writing:

A Plague Tale: Requiem
Atomic Heart
Bright Memory: Infinite
Cyberpunk 2077
Conqueror's Blade
Dakar Desert Rally
Deliver Us Mars
Destroy All Humans! 2 - Reprobed
Dying Light 2
F1 22
FIST: Forged In Shadow Torch
Forza Horizon 5
Hitman: World of Assassination
Hogwarts Legacy
Jurassic World Evolution 2
Justice
Loopmancer
Marvel's Midnight Suns
Marvel’s Spider-Man: Miles Morales
Marvel's Spider-Man Remastered
Microsoft Flight Simulator
Mount & Blade II: Bannerlord
Need For Speed Unbound
Perish
Portal with RTX
Redfall
Returnal
Sackboy: A Big Adventure
The Witcher 3: Wild Hunt - Game of the Year Edition
Warhammer 40,000: Darktide
WRC Generations

More from the Micro Center Community:

Looking for more information about NVIDIA tech? We’ve got a 40 Series Breakdown article, as well as benchmarks for the GeForce RTX 4090, 4080, 4070 Ti, 4070, and 4060. And if you can’t find what you’re looking for, don’t hesitate to post a new discussion and the Community will be happy to help!

Rye_Bread · June 2023

Great writeup! Lotsa useful information in here.

Ian · June 2023

Good information to know!

ChaosDTV · June 2023

good info!

McMuffin719 · June 2023

Pretty much everything you would need to know about DLSS, without getting too rabbit hole-y. Great read for anyone looking to narrow down a choice for a graphics card.

Vladdy · June 2023

Thanks for the information! We get this question often.

-V

ricardo_ortega · June 2023

Great way to summarize the details of DLSS’s features and advantages of it while keeping it easy to implement it. Thanks for providing this information!