If I ever had any doubts, I will definitely go for 60fps.
This is super misleading to say. It doesnāt run at any of those resolutions you listed. It reconstructs the image to 1080p in all scenarios (not upscaling, reconstruction). The actual IQ is fine in all modes on both next gen machines. Iāve played it on both.
Donāt know how many times it needs to be said but we are in a post resolution era where native resolution has very little meaning, because a lot of games use some form of reconstruction to reach higher resolutions from a lower base resolution.
The only thing that people should be concerned is if the final image is good and not what the ānativeā resolution is.
Not to mention the game is still very much in preview with a 2 months old build.
The temporal upscaling is great in Halo Infinte and fools gamer to be believe itās native.
Sometimes thatās better gives them more resources for other areas
The internal resolutions are same as the DRS resolutions.
Itās pretty simpleā¦
Internal resolution is DRS based from native to whatever lowest itās going. 1800p on horizontal axis on series X maybe.
The output resolution is always 4k. So sometimes it upscaled from 1800p to 4k and sometimes from 4k to 4k ( no upscaling I mean)
The major difference is the upscaling. Temporal upscaling takes the data from past frames to produce an upscaled image. Which may not be as good as the true output but far better then the true internal resolution.
Edit:
Temporal upscaling is a key here. It looks indistinguishable when frame is static. And while in motion at 60fps no human can detect itās artificats by naked eye. So itās always a win( well most of the time)
This should help you to understand the technology
Yeah I wasnāt trying to imply you were somehow dumb or something. The vast majority of ppl on forums have spent all last gen being conditioned to think internal res was all that mattered bc it was tied directly to image quality. That is no longer the case. It used to be that low res meant tons of jaggies, pixel flicker and muddy image quality in general.
What reconstruction techniques do (there are a variety of themā¦checkerboard rendering, temporal injection like UE5ās TSR, any ML super res soln like DLSS, etc) is look at multiple frames in a sequence prior to the current frame and use that info to guess what info is missing. That missing info refers to the way the methods scale the image by spreading the pixels out from their neighbors (as oppose to just stretching all the pixels. This leaves gaps that need to get filled in by the reconstruction algorithm.
The end result is the pixel density (which is what dictates how sharp/clear an image on screen is) is 4k. That said, these algorithms are still just guessing, so depending on the technique you will have different artifacts when that guess is slightly off from what it should have been. Thatās why when talking about just the internal rendering resolution it is super misleading, since the actual output image on screen is what matters and even with low internal res you can see great clarity on screen.
Complicating things a bit, sometimes you can have games use dynamic internal res and then reconstruction afterwards. Even more complicated, sometimes you can have something like Returnal that renders internally at 1080p, then uses a temporal reconstruction method (which works horribly for that game since almost the entire scene is in motion in every frame so there are artifacts everywhere), THEN uses CBR to get to a 4k output. That results in CBR taking in already-artifacted image and it can lead to the output frame on screen looking like a straight 1080p scaled to 4k in large sectors of the screen.
Then we have VRS on Xbox consoles, which changes the pixel rendering density based on where those pixels are on screen (if all neighboring pixels are basically the same color, no need to compute em allā¦just use VRS there to save on compute power). That can really confuse ppl since they are used to pouring over a screen shot that is static, can make areas with notable VRS application seems more notable than they are while actually playing (devs make sure this lowered rendering quality is only on areas of the screen players are not looking at).
Thereās a lot of complexity in general in the topic, which is why DF and @KageMaru and many of the rest of us have tried cautioning ppl about trying to boil image quality down to just a single internal resolution metric.
It means that at the most intense scenes, XSS internally renders 540p image, then uses temporal injection to reconstruct the rest of the pixels to get to 1080p output. Also important to note that we dunno if the 540p figure is actually applicable in both x/y directions here. Usually pixel counters only use one of the dimensions to count and some reconstruction approaches can be anisotropic (i.e. act differently in x and y directions).
As an example of where some approaches can shine over others, look at FlightSim. FlightSim only ever has very slow moving imagery when flying at normal altitudes (even zipping over rooftops is relatively slow moving compared to what temporal methods need for artifacts to start showing up). So it is the ideal choice for Asobo to go with since the only downside of temporal injection is the artifacts around the edges of certain fast moving objects, which are basically not present at all in the game while playing it normally.
Note: This is an image showing how well DLSS can do using a 540p internal render to target a 1080p output. DLSS is top dawg wrt reconstruction atm but other solns offer somewhat similar image quality results (the big difference settings DLSS apart is framerate perf). If I asked ppl which is native and which is āupscaledā without the labels I bet 90% of ppl would get it wrong!
But this is the resolution in a 120fps game, not some last gen 30fps stutter fest. When the time between two frames is only 8ms and not 33ms you get way better temporal stability. This means pixels between moving frames are only a quarter distance apart from 33ms and modern temporal reconstruction algorithms work much better with that (less errors). Thats why these low res games are now much less noticable than a 540p game like you sometimes have on Switch.
I donāt think itās that misleading. The temporal reconstruction falls apart kind of easily when in motion. Not watching the other videos until Iām done with my video today but Iāve counted a resolution as low as 540p on the Series S. The game is probablyā¦serviceable? on a 1080p display for the most part, but even on a 24" 1440p screen, it looked very blurry. Pretty sure Iāve expressed how low rez it looks in one of our times playing over the weekend. Itās especially rough on a 4K display. The 60Hz mode looks noticeably sharper hovering around 1080p but the lower resolution scaling is very noticeable, temporal reconstruction or not.
Edit: I also donāt think itās accurate or fair to compare DLSS to any other reconstruction technique. That gives a wrong impression when DLSS is leagues above anything else out there.
Only fast motion and even then not always. Many games donāt have fast enough movement on screen to lead to all that many artifacts and all scaling approaches will have some artifacting anyhow. Importantly, the IQ on XSS is nothing remotely approaching what youād get from a 540p output image w/bilinear upscale, which is what the poster was suggesting.
It doesnāt require that fast of motion though and thatās my point. Itās a very clear low, low rez IQ. Itās pretty bad and looks exactly like what Iād expect from a resolution that low. Did you play it at 120Hz?
