Eric Hu

"Noise is not acceptable" is the argument. When SSR came out, there were a lot of complaints about how noisy it was. Over the years denoisers for SSR became so good - you'll be hard-pressed to notice any noise today. When SSAO first came out - it looked quite shimmery and noisy, and there were a bunch of papers around that time, tackling the noise. Today SSAO still has noise, but denoisers take care of most of it. What's the most common anti-aliasing technique today? It's Tempotal-Anti-Aliasing, you might say: "No, actually it's DLSS/FSR", and I hate to break it to you - but those are temporal techniques as well. Yes they do a lot of clever stuff under the hood to reconstruct the image... based on previous frames, so it's still a temporal technique. Why does this matter? TAA is a kind of denoiser. We not only "accept" the noise, but the market clearly doesn't care as much as a vocal minority seems to indicate. The noise itself is a topic worth addressing. There are different kinds of noise, some is more noticeable than other. Today we heavily rely on diffuse noise, where the error is distributed uniformly, with blue noise dominating the industry. We know we're going to have noise - so we create the kind of noise that's easiest to denoise and that's least obvious to the human eye. Does it work 100% - no. Can you find failure cases - easily. But if we accept that people vote with their wallets - the people have voted, and it is in favor of the noise. Going further, If you believe that in raster ara we had no noise - your definition of noise is too narrow. Aliasing is also a type of noise, we undersample textures and geometry during rasterization, which results in aliasing artifacts, i.e. - noise. We've had noise for as long as we've had raster graphics. To preempt the most common rebuttal - I am not saying that noise is good, or that it is desirable. It is a byproduct of the techniques we use, and my argument is that the benefits that come with those techniques are outweighed by the side-effect of getting noise.