May i ask how do you train pssr? How does ML work? Is it learning by us playing more and sending some data back to the devs or is it per game?
Seems ridiculous we have to wait months to a year to have these games run with better IQ. At the very least i want the base modes back to bypass this crap till then
nvidia had some clips on youtube about this. yes, it comes down to having all kinds of different games being played on their AI servers over and over again, and eventually the AI trains itself and gets better.
DLSS is still getting better and better. On PC, we had to manually change the dlss version to get rid of ghosting on leaves in silent hill 2 because the game shipped with an older version of dlss. even avatar had this ghosting issue on crows and other flying objects so its been a thing for a while. death stranding was the first time i personally noticed it and that game came out in 2020. took 4 years for nvidia to fix it. so even dlss is continuously improving.
now the problem is that it took DLSS 6 years and PSSR has had maybe a few months or maybe a year of training? devs can also manually adjust and improve the algorithm. much like gg did with the checkerboarding tech. So i guess the answer is both. you can feed it data and it will also learn on its own. Maybe the can improve the algorithm by learning from Nvidia and AMD's mistakes.
For now, devs should not be using anything other than 4k pssr quality which uses 1440p internal resolution. Demon Souls, TLOU1, and TLOU2 use it and its fine. Some minor flickering in TLOU1 but its rare. Nothing distracting like spiderman 2 which uses 1080p to reconstruct to 4k or alan wake 2 which uses 1296p. though AW2's issues are mostly due to ray tracing not being handled properly so 1296p should be fine with PSSR. Nothing offensive like we are seeing in silent hill2 and jedi survivor.
Anything below 1080p should be banned until Sony can train it a bit more.