This yr at CES, Nvidia introduced the subsequent technology of its DLSS upscaling know-how, which is skilled with the assistance of synthetic intelligence, alongside the brand new GeForce RTX 5090, 5080, and 5070 (Ti) graphics playing cards. The corporate touted its main benefits — and now that RTX 5090 evaluations are reside, we are able to affirm that DLSS 4 certainly appears like black magic, supercharging body charges and making video games really feel simply as snappy because the beloved Doom 2016.
That’s as a result of DLSS 4 now helps Multi Body Technology (MFG), an AI-based a number of intermediate body calculation that may artificially generate as much as three photos and insert them between two “real” frames, thus quadrupling the body price. After all, this function solely works on new Blackwell-based RTX 50-series GPUs.
However are the AI frames generated on this approach a step ahead or is all of it hogwash? Let’s take an in depth take a look at DLSS 4 and its multi-frame technology works together with some early impressions.
Nvidia DLSS 4 intimately
The two-slot RTX 5090 Founders Version (proper) subsequent to the 3-slot RTX 4090.
Adam Patrick Murray / Foundry
Nvidia DLSS 4 builds atop the prevailing DLSS 3 and DLSS 3.5 function set. It’s made up of the next 5 capabilities:
- DLSS Deep Studying Anti-Aliasing (“DLAA”)
- DLSS Multi Body Technology (“MFG”)
- DLSS Ray Reconstruction (“RR”)
- DLSS Body Technology (“FG”)
- DLSS Tremendous Decision (“SR”)
The traditional upscaling know-how DLSS Tremendous Decision renders video games a decrease inside decision, then upscales to a better output decision with the assistance of AI. Tremendous Decision additionally works on older graphics playing cards from the GeForce RTX 20-, 30-, and 40-series, however it will get extra sophisticated with the opposite DLSS options.
Whereas the straightforward DLSS Body Technology (FG) is reserved for the GeForce RTX 40- and 50-series graphics playing cards, the newest DLSS Multi Body Technology function is just supported by the newest technology. DLSS Ray Reconstruction, which improves visible constancy in ray traced video games, is mastered by all RTX graphics processors.
Nvidia itself graphically summarized the complete function set of DLSS 4 by technology.
Nvidia
Along with DLSS 4, Nvidia can also be introducing a brand new Transformer AI mannequin, which replaces the beforehand used Convolutional Neural Community (“CNN”) because the neural community behind DLSS. Nvidia guarantees avid gamers even higher upscaling, extra exact ray reconstruction, refined AI edge smoothing (DLAA), and extra efficiency with the change. We will affirm it appears nice in motion.
The superior structure of the brand new neural community makes use of the precept of “Deep Attention” in addition to the worldwide context to generate considerably sharper particulars and considerably cut back artifacts equivalent to ghosting.
Nvidia
What’s Multi Body Technology?
Nvidia DLSS 4 now helps Multi Body Technology, an intermediate picture calculation that may artificially generate as much as three AI-calculated photos and insert them between two classically rendered photos. Commonplace DLSS 3 Body Technology solely inserts a single AI body between rendered photos. Nvidia exhibits the distinction in two easy-to-understand diagrams under:
The brand new know-how makes use of a number of neural networks, in addition to the Nvidia Optical Stream Accelerator and the optimized tensor computing items inside the brand new Blackwell GPU structure powering the RTX 50-series. Word that Multi Body Technology will not be backwards suitable with prior RTX generations.
Nvidia guarantees enormous leaps in efficiency via the usage of DLSS 4 with Multi Body Technology and makes this clear utilizing the instance of the notably performance-hungry motion role-playing recreation Cyberpunk 2077.
Adam Patrick Murray / Foundry
In PCWorld’s testing, each visible smoothness and responsiveness skyrocketed after flipping on DLSS 4’s Multi Body Technology in Cyberpunk 2077 (utilizing the RT Overdrive preset and 1.7x DLSS scaling). With none type of Body Technology on, the sport runs at 71fps; enabling DLSS 4 MFG lets it run a fully staggering 251 % sooner, and a just-as-insane 91 % sooner than the 5090 with DLSS 3’s single Body Gen energetic. Hitting such excessive speeds appears and feels so good.
However what impact do these so-called “fake frames” have on the gaming expertise and picture high quality?
DLSS 4: Is it true progress or “fake frames”?
Players who use Multi Body Technology have to be conscious that when it’s on, “artificial” frames are added to every “real” rendered body. However is that this in the end actual progress, as Nvidia CEO Jensen Huang described the function at his contentious CES 2025 keynote, or is it simply “fake frames” as some lovers have been shouting from the rooftops in on-line boards?
After all, the entire thing must be thought of in a differentiated approach and in unbiased exams. It’s comprehensible that patrons of a $1,999 graphics card need considerably higher efficiency with out “tricks” in comparison with the earlier technology. (Luckily, the RTX 5090 delivers stable conventional efficiency beneficial properties too.) However because the calendar flipped over to 2025, PCWorld declared that AI upscaling killed native graphics gaming. We’re higher off for it.
After utilizing it, DLSS 4 feels prefer it drives that principle house. Whether or not DLSS 4 provides true additional “frames” within the conventional sense or acts extra like movement smoothing on steroids is a vital technical distinction that calls for additional dialogue and testing. However there’s little question that video games feel and appear a lot smoother and snappier with DLSS 4 MFG and Nvidia Reflex energetic. It offers an enormous increase to the standard of your gaming experiences, interval (although I’d nonetheless hesitate to show it on in aggressive multiplayer video games).
We’ve got extra DLSS 4 protection coming, however right here’s a tease from our RTX 5090 evaluate: Turning on DLSS 4 makes Star Wars Outlaws, a enjoyable recreation liable to efficiency considerations, really feel simply pretty much as good because the legendary Doom 2016, which many avid gamers take into account the paragon of fast-action shooters. “It’s like a whole new game,” PCWorld’s Will Smith says. It’s excessive reward coming from a man who has been reviewing graphics playing cards and making video games for many years.
An preliminary conclusion
Using synthetic intelligence and options based mostly on it (like DLSS 4) ought to be welcomed if, for instance, clever AI upscaling improves each the image high quality and the body price — like DLSS Tremendous Decision does. The clever and intensely highly effective AI edge smoothing utilizing DLAA can also be a acquire, particularly for lovers.
The truth that Nvidia now not compares its new technology of graphics playing cards with native rendering efficiency from the earlier technology, as an alternative advertising considerably “embellished” body price values with lengthy bar charts in benchmarks, isn’t very useful. Comparisons of a GeForce RTX 4090 with DLSS 3 and Body Technology in opposition to a GeForce RTX 5090 with DLSS 4 and Multi Body Technology say completely nothing in regards to the precise enhance in rendering efficiency and are a supply of uncertainty for patrons. The poor preliminary reception by avid gamers is comprehensible and justified.
That being mentioned, now that we’ve examined it, DLSS 4 really is a revolutionary, game-changing know-how. Making an attempt to point out such a huge experiential leap ahead is difficult to do with easy body price graphs. I don’t like that Nvidia failed to point out how the RTX 5090 compares to the 4090 in conventional efficiency throughout its blockbuster CES keynote, however it’s considerably comprehensible now that we’ve skilled the distinction with our personal eyes and palms. DLSS 4 is that good.
Maybe the larger takeaway right here? {Hardware} and software-side AI options equivalent to upscaling, body technology, latency improvers, and devoted AI {hardware} are right here to remain and can play an excellent higher function sooner or later. The times of native rendering are coming to an finish. You definitely don’t have to love that, however I believe lovers want to start out accepting it. The sport has modified.
This text initially appeared on our sister publication PC-WELT and was translated and localized from German.
