Interestingly, NVIDIA, AMD, and Intel are all firing on screen zoom technology.
NVIDIA pioneered the creation of DLSS (Deep Learning Supersampling), which has become an industry benchmark, and has launched AI-based DSR (Dynamic Super Resolution), which has also enhanced NIS (Picture Scaling) and open source.
AMD has targeted FSR and RSR two super-resolution technologies, the latter of which is about to be launched, integrated in the driver, supports any game, and claims to improve performance by up to 70%.
Intel Xe architecture is ready for XeSS, claiming to be supported by more and more developers and games, and is expected to be officially launched by the end of the first quarter.
Unexpectedly, NVIDIA today announced a new "DLDSR," also known as "Deep Learning Dynamic Super Resolution."
Simply put, it is to add AI capabilities to the DSR, supported by the tensor core of the NVIDIA GPU on the hardware.
This means that it only supports RTX 20 and RTX 30 series graphics cards.
Although they are all scaling techniques, the direction of effort is different: DLSS, FSR, RSR, XeSS is a low-resolution rendering, and then increased to the screen native resolution output, DSR, DLDSR is the opposite, high-resolution rendering, and then compressed to the screen native resolution output, can greatly enhance the detail of the picture, make the edges smoother (less aliased), but also reduce glare.
After DLDSR joins the AI network layer, it requires fewer input pixels, and the image quality of DLDSR 2.25x is officially comparable to DSR 4x, and there is a higher frame rate.
For example, in the "Prey" game, the original performance is 145FPS at 1080p resolution, DSR 4x scales from 4K to 1080p frame rate to 108FPS, and DLDSR 2.25x scales from 1620p to 1080p to run 143FPS.
NVIDIA claims that DLDSR also does not require special optimizations for games, and most can be supported directly.
DLDSR will also be integrated directly into the driver, and the first version will be released on the 14th.