Nvidia has recently unveiled DLSS 3.5 which brings a wealth of new changes to the RTX functionality on all supported cards from the past three GPU generations.
Nvidia has revealed DLSS 3.5 which looks to change the way that RTX functionality works on all supported graphics cards when it launches in the fall. The big innovation is Ray Reconstruction, which further utilizes a card’s Tensor cores to push AI to improve ray tracing performance. In brief, it should mean ray-traced games will look and play better – and you won’t necessarily have to shell out for one of the best graphics cards to benefit.
I’ve argued before that ray tracing isn’t worth the performance hit and that’s because of how demanding the real-time lighting technique is on hardware, but this narrative looks to be changing with DLSS 3.5. It all comes down to how Nvidia’s ray tracing has worked up until now, utilizing the simulated light rays combined with a de-noising algorithm and upscaling to deliver a balance between visual fidelity and performance. That changes with DLSS 3.5 as Ray Reconstruction replaces the old de-noising technology with an algorithm built to fill in the image gaps more effectively and efficiently.
Simply put, Ray Reconstruction is a new AI model built by Nvidia that actively learns from supported software to more accurately display rays the way that traditional denoisers do not. It means that you lose less overall image quality when upscaling the picture from a native resolution to a target one, such as going from 1440p to 2160p without the massive performance hit. However, while DLSS 3 was previously only available on the RTX 40-series, this new update means that all GPUs from the RTX 20-series (2018), 30-series (2020), and 40-series are supported.
