sc94597 said:
Again, you are confusing two very different technologies. FSR 3/DLSS 3 aim to interpolate frames (add new frames where they don't exist already.) In order to do that well you need to have a decent frame-rate in the first place. There is also a significant performance cost associated with it. FSR 2/DLSS 2, on the other-hand, are temporal upscaling solutions. They're not interpolating frames, but rather pixels. In the case of DLSS this is by utilizing a convolutional auto-encoder that they've trained to infer pixels from past frame-data. AMD's FSR uses a more traditional method which doesn't use Deep Learning (many-layered neural networks.) The performance cost is relatively minor, in comparison to DLSS 3.0 (and ostensibly FSR 3.0), because in the case of an Nvidia RTX GPU there are specialized tensor cores designed to do it quite efficiently. Frame-rate improves with DLSS 2 because you run the game at a lower internal resolution, reducing VRAM demands and compute core demands, not because you are interpolating frames. This is the primary method through which people improve their framerates using DLSS/FSR. The consoles aren't going to get frame-generation. That doesn't mean that DLSS/FSR are useless. Starfield is a primary example of this. Without FSR 2.0, the Series X would run the game at an effective 1440p 30fps, and the Series S at an effective 900p 30fps. Instead, they are upscaled to 4k and 1440p respectively, and look somewhere in-between the internal and target resolutions in terms of image quality. PS5 titles (especially exclusive ones) often opt for a different temporal interpolator (most often checkerboarding) but the PS5 does support FSR 2.0 too. The idea that DLSS won't help the Switch in terms of performance or meeting graphical fidelity goals is just pure delusion. Temporal interpolators/upscaling solutions, which DLSS is the most successful implementation of currently on the market, are going to be (and arguably already are the) standard for the industry -- on both consoles and PCs. |
DLSS upscaling has a compute time cost that needs to be factored into the frame time budget but no input lag cost. Switch 2 probably need 100% of it's power to run ps5 games in 2 years lol. That screen also looks like crap in all 3 shots.
DLSS frame generation does has an input lag cost (as you need to render one frame ahead to interpolate the middle frame) but Switch 2 is unlikely to use frame generation, so that's a moot point.
