My point was mostly that even relatively high resolutions around 1440p can look quite bad depending on the upscaling tech/game. So if SX on top of having more performance can reach a point where the base resolution is lower the gap will get even bigger.
Though non ML reconstruction is bounded to improve too. UE5 is already reconstruction from 1080p to 4k without ML so we’ll need to see.
Regarding the quality of the ML solution Ms uses. I think it is more tied to the training data, and I would say Ms has an edge there because being a platform holder they can use all titles available on their platform for those sort of things (and they have in the past to fine tune the 1x and SX performance for example), but even if it doesn’t, it might not matter because they are aiming at a TV rather than a monitor you are going to sit super close so artifacts may be less noticeable.
Though Ms has researches in ML image reconstruction aiming to the same 64x multiplier Nvidia is talking about as the end goal for dlss.
I believe it will be before the current generation ends. By the halfway point, Sony will have ported the main games to PC and will then start day one on PC.
Its free money and we all make fun of Jim Ryan for his outrageous lying at times but I do believe he has done good business for Sony with the cross gen launches and PC porting.
I would argue that we are. It differs from game to game and developer to developer, but it’s more common than not to see a 10-20% resolution advantage on Series X that is entirely in line with the ~13% GPU delta between the two platforms.
It’s one thing to argue that, once you are past the 1080p mark, with modern reconstruction techniques that this doesn’t result a 10-20% IQ delta – that’s an argument I have some sympathy for – but this doesn’t mean you therefore get to strike that advantage from the record when it comes to comparing the power of the platforms objectively.
The difference is there. Not in every game, but certainly in most. We can stroke our chins over what it’s impact is, but we can’t just memory hole it and claim the systems have parity based on framerate alone. Or even more absurdly, claim the PS5 has a power advantage because it drops a frame once every half hour rather than twice – when 99% of the time both are running at 60fps and the Series X at a higher resolution.
I heard someone state that RT for reflections in racing games is incredibly computationally heavy due to high reflective surfaces on cars and fast movement meaning lots of calculations needed to make it look right. And the pay off may not be worth it.
To be fair, Playground made it clear from the start that RT for graphics was being used only in autovista.
But we are seeing that because RTing for a bunch of high quality cars are not a solved problem at the moment. The base geometry is too high for current RT hardware and using simplified models that make it performant apparently can make some stuff (mostly reflections I assume) weird. So they use in select modes.
