One of the staple features of NVIDIA's recently launched GeForce RTX series is something called Deep Learning Super Sampling (DLSS). It promises to deliver better performance in games that support it, though many have noticed that images tend to look blurry at lower resolutions when DLSS is turned on. What gives? NVIDIA addressed that question in a recent Q&A article, and also touched on a number of other related topics.
If you are curious about DLSS, the Q&A is a good place to start (hit the link in the Via field below). Andrew Edelsten, Technical Director of Deep Learning at NVIDIA, discusses what DLSS is and how it works. In short, it extracts large samples of aliased frames from games, and then for each one it generates a matching "perfect frame" using either super sampling or accumulation rendering. Those frames are then fed to NVIDIA's supercomputer, which in turn trains the DLSS model to recognize aliased inputs and generate high quality anti-aliased images that match the perfect frames as closely as possible.
"We then repeat the process, but this time we train the model to generate additional pixels rather than applying AA. This has the effect of increasing the resolution of the input. Combining both techniques enables the GPU to render the full monitor resolution at higher frame rates," Edelsten explains.
Not all games are the same, though, and depending on the game engine, complexity of content, and time spent on training, the results can (and do) vary. Edelsten also explained that DLSS is designed to boost framerates at high GPU workloads. As part of that, NVIDIA's initial focus has been on higher resolutions, and particularly 4K.
There is also an inherent weakness when working with lower resolutions. At 4K, DLSS has around 3.5-5.5 million pixels to generate the final frame, versus only 1-1.5 million pixels for 1080p. Having fewer pixels presents a bigger challenge for DLSS, hence why images look blurry.
"We have seen the screenshots and are listening to the community’s feedback about DLSS at lower resolutions, and are focusing on it as a top priority. We are adding more training data and some new techniques to improve quality, and will continue to train the deep neural network so that it improves over time," Edelsten added.
These improvements are pushed out through driver updates, so if you own a GeForce RTX card, you'll want to keep an eye out for those. How big of a difference it ultimately makes remains to be seen, and you can count AMD among the skeptics about the technology as a whole. During a pre-launch briefing for its Radeon VII graphics card, AMD threw some shade at DLSS.
As far as AMD is concerned, open standards such as SMAA and TAA don't come with "the image artifacts caused by the upscaling and harsh sharpening of DLSS," PCGamesN reports.
Incidentally, Edelsten's Q&A article touches on the topic of TAA as well. He points out that TAA "can suffer from high motion ghosting and flickering that DLSS tends to handle better," though does concede that depending on the resolution, quality settings, and game implementation, some gamers prefer TAA in one game and DLSS in another.