published on

Liabe Brave (528482)

Title: All games with PS4 Pro enhancements

Red Text: outlined in red

That is the conclusion I came to and posted back there, but I was wrong. I have no excuse for the error, because these are some of the exact same screenshots I analyzed then. It appears I only counted a couple edges, likely because these shots were messy, imgur-compressed JPEGs, and this colored my analysis of cleaner PNG shots. , thank you very much for resurrecting the topic and (unwittingly) giving me a chance to reverse my mistake. Say we start with this shot: So, the area is an obvious candidate for counting: a long edge obviously intended to be straight, with high contrast. (The window divider to the right is even longer, but also surrounded by worse JPEG noise.) Looking at the edge (200% zoom), each step is exactly 2 pixels wide and ratio count thus gives a 2x upscale result. I must've ignored the horizontal lines since they aren't really countable, being completely flat with no steps at all. However, this made me miss a crucial observation. Look at the area , again zoomed 200%. It's instantly apparent that some horizontal lines are only one pixel tall. This is literally impossible in an upscaled 1080p image. So we then go to another image of the same environment, this time from an angled camera view. The near-vertical lines still count as 2x upscale, but the near-horizontal ones count as unscaled: one pixel for every step taken. (You can still see some "chalkmarks" from my counting, partially eliminated by a nearest-neighbor downscale.) Because of what pixel counting is actually measuring, near-vertical lines are a measure of resolution width, and near-horizontal lines a measure of resolution height. Why the strange number? There are a few possibilities. First, the game could be running an anamorphic dynamic resolution, where only the width changes. This isn't an uncommon approach ( is using it), but the facts seem against it here. In such games, an image less wide than tall is usually only seen at the very lowest drops, when the engine can barely keep up with the action. That definitely doesn't seem likely in these scenes, especially since they vary in content but all show evidence of this sizing. Second, the buffer might simply be static, and upscaled in only one direction. But while this is not unheard of, choosing to use full height when that will restrict you all the way to half width is unusual. Such a 1920x2160 buffer is more total pixels than full 1440p, and only a little bit less than full 1620p. Given the more balanced distribution of increased-res benefits, those other sizes would seem more logical choices if you're restricted to this level of throughput. Third, some temporal reconstruction methods (CBR, TAA, etc.) may use undersized buffers at some point in the process. These are definitely not supposed to end up being output to the screen, they're just the building blocks that the reconstruction operates on. However, we have precedent for something like this sneaking through. When it launched, also rendered taller than wide, to much confusion. A patch later eliminated this, and the final version doesn't seem to use reconstruction at all; but could this be an example of initially botched reconstruction that might guide us with ? Well, there may be artifacts present in another of the images (400% zoom). Notice the light/dark/light/dark bands near some edges. This could just be mosquito noise from JPEG compression. After all, the large block impressions here are certainly JPEG color channel chunking, and there's some less-structured noise elsewhere in the zoom. (Not to mention the blending from the anamorphic upscale.) But these are strong, exact alternations at pixel scale, so I suspect they may be part of the original image. There are also more--though even less certain--suggestions of temporal artifacting in other shots. And tellingly, the impressions from the GameFAQs thread say they're visible if you stand close to a 4K screen. Unfortunately, the only 4K PNGs I have were taken during little movement, so are inconclusive; there seem to maybe be artifacts, but it could just be specular aliasing, etc. Until I get better captures where stuff was in heavy motion, I can't conclude exactly what the reason behind the resolution is. Maybe there's no reconstruction at all. But at least I can now say that have Pro support, though it's not of an expected type. This might not be intentional (though the amount of bloom and DOF applied to this game might be evidence otherwise, of an effort to efface the worst artifacting of the chosen resolution). My apologies once more for the previous error, and any confusion it may have caused.

Link: https://www.resetera.com/posts/528482/

Archive.org: https://web.archive.org/web/20180709070301/https://www.resetera.com/threads/all-games-with-ps4-pro-enhancements.3101/page-4

Archive.is (legacy): https://archive.is/nIcHr