A look at how NVIDIA’s DLSS 4 and Reflex 2 technologies work

DLSS 4 technology brings with it the new Transformer model and the Multi Frame Generation technology that generates more intermediate frames, which is exclusive to RTX 50 graphics cards, while Reflex 2 borrows the playbook from the virtual reality side.

At the CES 2025, NVIDIA released not only the new GeForce RTX 50 graphics cards of the Blackwell architecture. Alongside the new graphics cards, the new DLSS 4 and Reflex 2 technologies were also launched.

According to NVIDIA, DLSS technology is already used in more than 540 games and applications. According to the company’s statistics, even more than 80% of the users of the company’s RTX graphics cards also use DLSS technology, at least to try it out.

With DLSS 4, the technology is renewed in many ways. Until now, both scaler and ray reconstruction have been based on the CNN (Convolutional Neural Network) model, but DLSS 4 brings a new Transformer model. The new model enables double the number of parameters, but is also four times heavier than before. The new model should offer clearly better quality in certain situations. The Transformer model, which has increased in popularity especially in artificial intelligence image creation in recent years, is supported by all RTX graphics cards.

Another significant innovation in DLSS 4 is the Multi Frame Generation technology, which, unlike other updates, is only available with GeForce RTX 50 series graphics cards. NVIDIA has replaced the use of the Optical Flow Generator unit in the Blackwell architecture with an artificial intelligence model and added support for Flip Metering technology, which enables better pacing of screens than before. Multi Framew Generation practically means that instead of one added frame, the graphics card creates three intermediate frames between the new frames. If and when DLSS scaling is also used in Performance mode, only one of every 16 pixels is rendered to the traditional model.

Along with the four generated frames of DLSS 4, NVIDIA’s new lag-reducing Reflex 2 technology is also essential. The new Reflex 2 brings to the PC world the familiar Frame Warping technology from the virtual reality side to enable the lowest possible delay from the user’s input to the screen. So in practice, first the screens are rendered as late as possible, as before, after which the last user inputs are read and the rendered screen is “warped” to match them. As a negative side of the technology, it can create errors, especially at the edges of the screen, where things that have not been rendered should appear, but apparently the screen update speed, especially with Multi Frame Generation, and the help of artificial intelligence in filling are enough to keep the problem at bay. The Reflex 2 game has to be built into the game separately, and so far at least The Finals and Valorant have signed up. The technology will be available on all RTX graphics cards, but initially it will be exclusive to the RTX 50 generation for an unknown period of time.

New DLSS Override settings will also be added to the NVIDIA App, which will make changing DLLs a thing of the past. Through the settings, it is possible to change the DLSS model used, the number of frames generated by Frame Generation and the DLSS Super Resolution settings, including the last option to use DLAA technology, globally.

Source: NVIDIA press materials

Source: www.io-tech.fi