The way DLSS3 and 3.5 work (though not DLSS4 it seems) is they always guess a previous frame, never a next frame. They always rely on an authentic frame and guess an intermediate frame between prior to it, because they need to know the end state and interpolate backwards from it.
You are not logged in. However you can subscribe from another Fediverse account, for example Lemmy or Mastodon. To do this, paste the following into the search field of your instance: [email protected]
Welcome to the largest gaming community on Lemmy! Discussion for all kinds of games. Video games, tabletop games, card games etc.
Yeah, turn based CRPGs is the only time I turn that stuff on.
But there seems to be two types of frame generation:
A real frame is generated, then it guesses, then real frame is generated.
Two real frames are generated, then they “fill in the gaps” with fake frames.
For 1 it doesn’t create lag, but it may guess wrong then have to correct on the next real frame.
For 2 the fake frames are going to be correct, but a lot of latency is added.
Neither is an issue for CRPG, but can be a huge issue in shooters, fighting games, and PvE like Dark Souls.
The problem is manufactures are now counting all the fake frames in their stats while ignoring every issue that pops up from that.
The way DLSS3 and 3.5 work (though not DLSS4 it seems) is they always guess a previous frame, never a next frame. They always rely on an authentic frame and guess an intermediate frame between prior to it, because they need to know the end state and interpolate backwards from it.
This is described here: https://en.wikipedia.org/wiki/Deep_learning_super_sampling#DLSS_3.0
I’m not sure how AMDs frame gen works, is it different?