Nvidia has revealed a new version of its DLSS (Deep Learning Super Sampling) upscaling tech, called DLSS 3.5.
This partial new new generation, the first for DLSS, adds one new tech to further improve the quality of ray-traced graphics: Ray Reconstruction.
Ray Reconstruction works better than Denoising techniques, which are used for filling missing pixels when upscaling ray-traced computer graphics. Nvidia has trained DLSS 3.5 with 5 times more training data than DLSS 3.5
So what’s the result of this? Ray-traced effects, like reflection from puddles, are slightly sharper. You really need to look at the comparison shots really, really closely to see the difference. Especially if you stop minding about blurry graphical effects and accept them as is all these years. Which to those folks, the effect is marginal at best.
DLSS 3 previously added Frame Generation, where the GPU creates new in-between frames using AI tech to further boost gaming performance, useful when playing games with high graphics settings but not wanting to sacrifice framerates.
Ray Reconstruction can be used on all Nvidia RTX GPU cards.
UPDATE 23/08/23: An earlier version of the article stated that Ray Reconstruction, the new tech for DLSS 3.5 was only available for Nvidia RTX 40 Series GPUs. This was an error in our reporting, and the article has been updated.