Gnnnnnnnnnx! It keeps searing my eyes, every time I see continued perpetuation of calling it: “supersampling”.
Apart from your eyes, is there anything else that bothers you about calling it supersampling?
Heh, I sense some sort of schooling incoming… :9
As I suspect you may gleefully be rubbing your hands in anticipation of my answering, possibly with a response just about fully prepared: To my mind, “super”-sampling would have to involve “for-real” sampling one’s projected view of the world at higher rate than the requested output bitmap size. Sampling it lower, and then upsampling the resulting image, no matter how cleverly… mneeeh…
It’s the same kind of feeling I have about 2x multisampling being two samples, rather than two samples in each dimension (I think I may have bickered with you about that one some time(s) before :9), or harddisks being sold in sizes of decimal kilos and megas, and so on, rather than the familiar binarily aligned approximating 1024, which I suppose should be kibis and mibis, etc…
It’s deceptive, and there is no lack of people who will barge right down any paths to misunderstanding that have been carefully set up by some sly individual in marketing, and come to actually believe and evangelise any thing that makes something sound infinitely better than it really is . :7
One could argue e.g. that DLSS does sample temporarlly, which adds some, and that both solutions demand well antialiased input, which could involve various sorts of extra detail being pulled in, in some way – heck, a mipmap alone could be said to constitute something along the lines, but… that’s at least one level removed, as far as I am concerned… :7
Ha ha, I had no intention of that at all, just was genuinely curious about the hint you made and that maybe I missed some substantial property (or a lack of) af the two aforementioned technologies. And I always enjoy your witted prose .
I see what you mean and I agree that the term does not really fit its subject, but I guess someone at the marketing thought that super sampling sounds cooler than up sampling (which would be used in the context of opposite of downsampling), while completely omitting the fact that there is no sampling (in a traditional sense), because from the already rendered image, you can only get as much “samples” as there are.
I do not consider the temporal behavior of DLSS (which btw seems to be more of a problem than an advantage) the supersampling either. Unless you call any filter which incorporates previous values a supersampler.
Oh… Maybe lessons next time. :7
A problem bigger than shimmering? In VR? I can’t believe it.
Anyway it’s impossible to separate DLSS scaling from temporal antialiasing, so I need to considered it a single package.
When I hope for DLSS implementation in a VR game I don’t think about the resolution increase or fps boost, I just think it will certainly have temporal antialiasing. And I love it.