AMD, Intel and NVIDIA all have their own super-sampling techniques designed to speed up graphics rendering. And what about image quality?
Since the very first version of Deep Learning Super Sampling (DLSS) imagined by NVIDIA and launched, in 2019, on Battlefield 5 in particular, it has been clear that these super-sampling techniques, now boosted by artificial intelligence, are likely to greatly simplify the task of our graphics card GPUs. Thanks firstly to powerful algorithms, and now to AI, NVIDIA's DLSS - but also AMD's FSR and Intel's XeSS, its main competitors - are able to display a game in 4K (2160p), while the GPU makes do with rendering in 1440p, or even 1080p. Better still, AI is now exploited to generate images "from scratch" between those actually rendered by the GPU. Inevitably, the gains in animation speed and image fluidity are enormous.
It's hard to decide whichHorizon Forbidden West image is the prettiest © ComputerBase
One question, however, has regularly entered the debate without being really settled: is the visual quality induced by these technologies sufficient and, more importantly, is it up to the level of what we'll call "native" rendering, in other words, entirely realized by the GPU? This is not a simple question to answer... and it's not our intention to do so today. On the other hand, by relaying the survey carried out by ComputerBase - a German site specializing in gaming hardware - we can provide some elements of an answer. ComputerBase played a little game: offering short video sequences based on some of the most popular games(Anno 117, ARC Raiders, Cyberpunk 2077, Horizon Forbidden West, Satisfactory and The Last of Us Part II). In each case, the games were offered in "native" mode with TLAA enabled, in DLSS 4.5 mode (1440p base) and in FSR Redstone mode (1440p base). Without knowing which video went with which technology, participants were asked to state their preferences.
Not so easy to choose on The Last of Us Part II © ComputerBase
The results are interesting because, out of a total of 6747 votes cast, NVIDIA technology (DLSS 4.5) still accounted for a total of 3249, or 48.2% of the votes. Far behind, "native" mode is at 24% (1619 votes) and FSR Redstone is third at 15% (1013 votes). It's worth noting that 12.8% of votes (866 votes) were cast in favor of "renderings are equivalent", proving that choosing is not so simple. Note also that XeSS 3 technology, released too recently, could not be included in this test. While DLSS 4.5 appears to be the clear winner, there are a few comments to be made, starting with the activation of TLAA for native mode: the technique is widely used to improve rendering, even in native mode, but it leads to a certain blurring that DLSS and FSR technologies are trained to combat. What's more, the 1000-day sample, while interesting, remains highly inadequate, and the selection of six games also has its limitations.
