It’s been about a month since Nvidia introduced another big version of DLSS, which looks more like a branding package of several technologies. In addition to scaling or frame generation, version 3.5 offers Ray Reconstruction (RR), that is, ray reconstruction. Until now, path tracing had to use several so-called denoisers (full ray tracing implemented for general lighting, reflections, shadows, background occlusion and, in short, all aspects). Algorithms that ensured that the resulting image could be used at all, because the “raw” one was full of noise. But the need to use multiple noise cancellers always meant that certain compromises had to be made. And this is exactly what Nvidia wants to solve with RR technology, which works as a single noise canceller, trained using artificial intelligence, and designed to replace all others, thereby simplifying the entire process. Also, unlike frame generation, it works on all RTX cards, i.e. 20, 30 and 40 series.
A few days ago, update 2.0 was released for the game Cyberpunk 2077, which, among other things, brought the reconstruction of beams. It only works with path tracing enabled (RT Overdrive mode) for now, and it’s unclear whether Nvidia will release it for more limited ray tracing in the future, which it says won’t be as useful. It should be borne in mind that the technology goes hand in hand with upscaling, so it can only be used with DLSS enabled. You can’t enable it at native resolution, even when using DLAA anti-aliasing. Again, this is something Nvidia may add in the future.
Of course, several tech-oriented magazines and channels have taken notice of RR. I would mention, for example, Hardware Unboxed, Daniel Owen, Gamers Nexus and, of course, Digital Foundry, where Alex Battaglia did some pretty detailed analysis. The result is very positive, thanks to RR the reflections are sharper and more accurate, the image is much more stable (without various flickers and flickers) and the shadows (mainly Ambient occlusion) are much more noticeable and realistic. Performance is often even slightly better than without RR, but this is not always the case and is not the main goal of this technology. However, it is important to note that this does not degrade performance in any way.
Unfortunately, there was some negativity. RR can cause relatively severe ghosting and general blurring of some objects, especially small ones, but can also affect characters, for example. It’s definitely not a pretty look, and we hope Nvidia improves it in some future version. Otherwise, almost everyone agrees that Ray Reconstruction significantly improves the image. By the way, Daniel Owen offers an interesting comparison in his video, which shows that RT Overdrive mode with DLSS Quality, Ray Reconstruction and Frame Generation technologies enabled offers frame rates similar to native resolution without ray tracing.
Source :Indian TV
