Yesterday, the NVIDIA RTX 40 series was officially released. I believe that while everyone feels that the price is still expensive, the multiple re-tech released by NVIDIA is also eye-catching. The biggest highlight is the optical multi-frame generation technology DLSS 3. Compared with actual game tests without this technology, the performance is directly improved by 4 times. In addition, since the DLSS generated frame is executed as a post-processing on the GPU, performance improvements can be obtained even if the game is limited by CPU performance.
DLSS 3's optical multi-frame generation technology revolutionizes the technology, which can directly generate a completely new full frame picture, not just a single pixel. It uses optical flow accelerator to analyze two consecutive game images and calculates the motion vector data of objects and elements between frames, rather than modeling using the motion vectors of traditional game engines, which greatly reduces the visual anomalies of AI when rendering elements such as particles, reflections, shadows, lighting, etc. In turn, create other high-quality frames without affecting image quality and response speed.
For the technical characteristics of DLSS 3, it is currently only exclusive to the RTX 40 series. However, NVIDIA technology senior management said: Optical flow accelerators are not newly added to the RTX 40 Ada architecture. They are actually all in the RTX 20 Turing and RTX 30 Ampere architectures, but they are not powerful enough. I believe this is just like the GTX 16 series can also support ray tracing technology, but because of the poor actual experience, users are unwilling to upgrade. Of course, if NVIDIA intends to use this technology to the RTX 30/20 series to benefit old users, it will rely on certain technical support and improvements.
Finally, let’s briefly review the history of these generations of DLSS:
The first generation of DLSS, which was born in the RTX 20 series, combines AI and resolution amplification technology for the first time to achieve the purpose of improving rendering performance. However, because the AI model is not perfect and some conditions are immature, the performance at that time was quite disappointing, especially the blurred picture.
DLSS 2.0 version has adjusted the technical ideas and greatly improved efficiency and processing speed. The official claims that its speed can reach twice that of the previous generation. DLSS 2.0 extends the multiple of supersampling, can support 4x resolution stretching, greatly saving GPU resources and providing higher frame counts. The
DLSS 2.x version has greatly reduced the threshold for developers to use, and has also improved many unfinished problems such as ghosts that have existed before, thereby further improving the game quality. It has now received support from more than 200 games and applications.
