In recent years, PC gaming has evolved enormously, changing paradigm regarding the performance of video cards, no longer obtained exclusively through brute power but also thanks to technologies. At the basis is the evolution of DLSS, which has profoundly changed compared to its origins.
In retracing the history of DLSS it is necessary to make a rough chronological distinction: for NVIDIA’s technology, in fact, there is a before and after version 3.0.
Il DLSSacronym for Deep Learning Super Sampling, was born as a technology capable of upscaling and upsampling images in real time. The “magic” of the green team’s supercomputers, in fact, can reconstruct images in an artificial way, generating the missing pixels in the transition from one resolution to another in a storm thanks to AI. A PC running a game with DLSS enabled therefore does a a lower resolution than what the player sees on the screen thanks to the potential of the Tensor Cores of NVIDIA GPUs. From the first to the second version, NVIDIA has worked both on the final rendering of the reconstruction and on facilitating its adoption by software houses, thanks to the from single-title supercomputer training to generic training sets, making implementation much easier.
In all subsequent version 2 updates, DLSS has had a mainly iterative evolution, improving libraries and performance. As we were saying, there is also an “after”. In fact, with the launch of the RTX Series 40, NVIDIA presented DLSS3initially without space between the name and the number to indicate not only its evolution compared to version 2.0, but also to highlight the transition from a single tool to a container of technologies, exactly 3:
SuperSampling
Frame Generation
NVIDIA Reflex
The first is the “original” upscaler, supported by the revolutionary Frame Generation which, thanks to Optical Flow technology, is capable of add reconstructed intermediate frames averaging the information from the previous frame, the next frame and the motion vectors. In the end, NVIDIA Reflexnecessary to lower latency due to the increase in latency resulting from the use of Frame Generation (which analyzes “subsequent” frames not yet on the screen, slowing down the rendering queue).
The result was an earthquake for the industry: a switch to double or even triple the framerate, without affecting the visual quality (if anything, some artifacts in the most frenetic scenes). Finally, launch of DLSS 3.5: in 2023 NVIDIA presented its latest effort, which adds a .5 to indicate the presence of a new technology, Ray Reconstruction. This technology focuses on improving the performance of Ray Tracing, retiring the old and poorly performing denoisers and using a new system based on Deep Learning, which is much more efficient and which uses processing and time to make the reconstruction even more faithful. of the image.
NVIDIA always has been very secretive about the future of DLSS, surprising with each launch of new concepts. The next pages are still to be written, but we are sure they will be exciting!