-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Fidelity Super Resolution in 2021
- Thread starter Grim5
- Start date
Gotta say, I have to walk back my positive impressions of FSR. Decided to play some more Riftbreaker last night and ended up seeing a LOT more problems with it when the camera pans left<->right (which is most of the game), and a lot more obvious at lower quality settings but definitely apparent even on FSR-UQ even compared to just reducing res scale to 75%. Thought, ok, maybe it's just a one off but today I finally give Godfall a try (and btw if you play this game you MUST enable FidelityFX LPM - on vs off it's like two different games, no joke, the difference visually is staggering) and have noticed FSR UQ not really standing up to just 1800p (vs 4K) either. The image is just too soft and the grain is accentuated in an already grainy game. No sharpening vs FSR it's ok, but as soon as you add sharpening to it immediately FSR falls apart in the comparison.
Gonna try see what I can tweak in the config files (TAAU et al) but for now I am starting to like FSR a lot less. I'd be interested to know if they can add the RCAS pass alone to the games, an improvement to just the sharpening is a lot more interesting right now.
Gonna try see what I can tweak in the config files (TAAU et al) but for now I am starting to like FSR a lot less. I'd be interested to know if they can add the RCAS pass alone to the games, an improvement to just the sharpening is a lot more interesting right now.
I'm seeing no such issues with FSR in the RiftBreaker and have been testing it a fair bit at 4K Ultra Quality on FSR. For me the only thing FSR introduces is a very small level of blur that I need to pixel peep to see. Are you using sharpening with native etc?
The only other game I have tried it in is Terminator Resistance and FSR actually improves the image quality there. But I suspect it's the low res textures getting improved with the sharpening pass in FSR. I played the whole game with FSR enabled and had absolutely no issues with motion or blurring etc.
The only other game I have tried it in is Terminator Resistance and FSR actually improves the image quality there. But I suspect it's the low res textures getting improved with the sharpening pass in FSR. I played the whole game with FSR enabled and had absolutely no issues with motion or blurring etc.
Yup, TAAU is WAY better in Godfall, just tested it now at 85%. It manages to retain that 4K sheen where you have all the specularity you'd normally expect, e.g. when the muddy ground has that wet look, FSR would just smooth it and completely kill the light glinting but it's retained with TAAU. It's almost HDR vs SDR like. Amusingly I am noticing more artifacts now, particularly for left<->right camera panning here as well, for example near metal grates if you move the camera quickly enough you can almost see a rainbow effect, or when you get to choose a new valor plate the effect happens quite obviously there too. Luckily it's not noticeable when you're just running through forward, or in combat. Didn't test just regular TAA for that to see what the difference is, but definitely clear to see with TAAU. Similar to what I saw with Riftbreaker in a way but there it was a different kind and much worse.
I would be curious to see if anyone can record some DLSS footage of moving the camera left<->right, perhaps Cyberpunk or WD:L, particularly near structures that make such artifacts more obvious (fences, metal grates, stairs, straight vertical lines etc.). I'll see if I can record for others.
I would be curious to see if anyone can record some DLSS footage of moving the camera left<->right, perhaps Cyberpunk or WD:L, particularly near structures that make such artifacts more obvious (fences, metal grates, stairs, straight vertical lines etc.). I'll see if I can record for others.
What problems do you see when panning? FSR only affects a single frame so there should be no issues during movement. If in Riftbreaker you are seeing a blur affect on the main character when moving then that is the game itself, not FSR. I've played it and see no issues at all on my 165Hz monitor so perhaps it's a display issue rather than the tech.
Also you seem to be comparing a 'sharpened' native image to FSR which is essentially moving the native goalpost. FSR is attempting to match native but it can't do that if the native image is further enhanceed with sharpening. Have you ever seen a DLSS comparison where they sharpen the native image??
here u go, dunno if thats the type of fotoage u wanted tho
https://www.youtube.com/watch?v=VmksbpEySvg&feature=youtu.be
i notice no flickering or artifacts with cp2077's dlss, only ghosting. actually, game's own taa seem to have more artifacts than dlss
(just a side note, i dont normally play with 1440p dlss quality, as you can see 3070 can't handle even that. it literally needs 720p input to push 55+ frames. sorry for the low framerate, that's the best this so called "rt capable ampere" gpu can deliver at its designated "1440p" target resolution. notice i even had to sacrifice volumetric quality setting to gain some more framez xd)
Last edited:
Caporegime
As above, Image Sharpening should be disabled in Radeon Software and or the game video menu (Fidelity FX Cas, or LPM etc) otherwise you are not comparing to a true native image.
I'll install it again some other time, the issue is one of image stability and it's not the same when using FSR vs no FSR. It's to do with the tech itself. The tests were done at 60 fps locked, if you go HFR then due to the nature of temporal AA you will be less likely to notice issues. The bit about sharpening I address in another response below.
Thanks, that looks good. The recording compression is off the charts with these tests I've noticed doing my own now, it's too bad. Here's what I meant before:
https://drive.google.com/file/d/1vmSBGwnwBMLgCTkYrJ8JaI9HVdgeh5cu/view?usp=sharing
Kinda hard to see until 4K is done processing but you can notice the blue lines behind the characters, I'm guessing the pixels pick up colour from the previous frame and hold on to it, so that's why they kinda go from light blue to dark blue as I rotate the camera. I tried to do TAA vs TAAU earlier and it seems like it's more obvious with TAAU but I can't swear on it just yet. Recording quality even at 100 MB/s starts compressing too much too quickly and changing between AA requires a game restart.
RIS is disabled, and LPM has nothing to do with sharpening or this discussion in particular, I only mentioned it as an aside (it's for HDR). Whether native has sharpening or not is entirely irrelevant because I can always add it so why would I care about what things look like without it. I see no reason why you wouldn't compare versus with-CAS-added except if you don't know the sharpening levels & then you're worried that you might overvalue the sharpening effect rather than something else. Rest assured I'm intimately familiar with it and have no issue discerning what's what when it comes to sharpening. Furthermore, I put 0 value into whether or not this is a "fair" comparison or what have you, the bottom line is simple: I'm comparing the image quality of two(+) options, the only thing I care about is the end result.
In particular whether it makes sense to use this at all. For Riftbreaker it's more complicated and I want to say FSR UQ is worth it, but it's in-development and the vsync is finnicky so it's a more complicated question that I can't settle atm. In DotA 2 the extra sharpening quality is welcome. In Godfall - HELL NO, it's a big downgrade imo because of losing that "sheen". For the rest we'll see, I haven't tried them.
And finally this is still work in progress, I am open to change my mind as I test it some more & discover more oddities. Right now though this is how I see it.
no worries, i actually tried TAAU in godfall waaay earlier. i just don't want to install that garbage anymore... but im inclined towards it nowI'll install it again some other time, the issue is one of image stability and it's not the same when using FSR vs no FSR. It's to do with the tech itself. The tests were done at 60 fps locked, if you go HFR then due to the nature of temporal AA you will be less likely to notice issues. The bit about sharpening I address in another response below.
Thanks, that looks good. The recording compression is off the charts with these tests I've noticed doing my own now, it's too bad. Here's what I meant before:
https://drive.google.com/file/d/1vmSBGwnwBMLgCTkYrJ8JaI9HVdgeh5cu/view?usp=sharing
Kinda hard to see until 4K is done processing but you can notice the blue lines behind the characters, I'm guessing the pixels pick up colour from the previous frame and hold on to it, so that's why they kinda go from light blue to dark blue as I rotate the camera. I tried to do TAA vs TAAU earlier and it seems like it's more obvious with TAAU but I can't swear on it just yet. Recording quality even at 100 MB/s starts compressing too much too quickly and changing between AA requires a game restart.
RIS is disabled, and LPM has nothing to do with sharpening or this discussion in particular, I only mentioned it as an aside (it's for HDR). Whether native has sharpening or not is entirely irrelevant because I can always add it so why would I care about what things look like without it. I see no reason why you wouldn't compare versus with-CAS-added except if you don't know the sharpening levels & then you're worried that you might overvalue the sharpening effect rather than something else. Rest assured I'm intimately familiar with it and have no issue discerning what's what when it comes to sharpening. Furthermore, I put 0 value into whether or not this is a "fair" comparison or what have you, the bottom line is simple: I'm comparing the image quality of two(+) options, the only thing I care about is the end result.
In particular whether it makes sense to use this at all. For Riftbreaker it's more complicated and I want to say FSR UQ is worth it, but it's in-development and the vsync is finnicky so it's a more complicated question that I can't settle atm. In DotA 2 the extra sharpening quality is welcome. In Godfall - HELL NO, it's a big downgrade imo because of losing that "sheen". For the rest we'll see, I haven't tried them.
And finally this is still work in progress, I am open to change my mind as I test it some more & discover more oddities. Right now though this is how I see it.
i found TAAU to be succesful, at least in terms of sharpness and clarity. but i also experienced the shimmering you've observed, its crazy. if you want to document it with more intensity, try 1080p. 540p internal resolution amplifies the shimmering problem even further
Interesting, I've not seen people do that before (super-sampling to 4k with DLSS and then back down to 1080p). Back in the days people just super-sampled up and scaled down but that costed a lot of FPS, so this is interesting. I imagine one could do identical thing with FSR, which gives me ideas!
Caporegime
Fair enough, just thought I'd check.
I posted my comparison in Godfall using true Native vs Ultra Quiality and when playing I can't really notice a difference unless I take screenshots and start pixel peeping at 4K with screenshots.
If pixel peeping then i could see a very slight reduction in image quality, softening of the image. There were no artifacts or issues with movement either when playing, other than a 50% increase in FPS.
A worthwhile trade for me given the graphical demands of Godfall max settings at 4K, but I can understand that others may see it differently.
FSR Native v Ultra Quality - Imgsli
DLSS works very very well on edges and lines - it's what it was designed to do more than anything and I am quite confident it could regenerate line from literally just 3 points (just drawing one crossing all 3 points). It doesn't work that well on textures though, as that is much harder to do with small number of input pixels - hence image becomes soft looking, unless you go to high resolutions and use Quality preset. Still, it looks good enough and increases FPS - this is what most gamers care about. Pixel peeping is fine to compare but has nothing to do with actual gaming.
Interesting, I've not seen people do that before (super-sampling to 4k with DLSS and then back down to 1080p). Back in the days people just super-sampled up and scaled down but that costed a lot of FPS, so this is interesting. I imagine one could do identical thing with FSR, which gives me ideas!
yup, this guy did some extensive stuff
https://www.reddit.com/r/Amd/comments/o61yc4/using_fsr_amd_virtual_super_resolution_or_nvidia/
supersampling these days do not only grant better aliasing, but they also provide sharper, more clear image. i really don't think any kind of native resolution actually provides native resolution image quality with TAA. the visual gains i see proves this, there's no way i should notice this much of a difference, yet i do, and that only means that TAA actually destroys a huge amount of pixel information and supersampling gives back that information. i think taa is clearly meant for native 4k rendering, its subpar for native 1440p, and outright horrible for native 1080p. hopefully supersamplign saves the day and dlss/fsr on top of that is nice to have
That is very not true. As in, just using logic - you can't create something out of nothing, hence if you're missing pixels dropping below 1080p then there's nothing to upscale from. Hence, in 1080p upscaling from 540p or even 720p FSR looks bad, DLSS (with AI) looks bad. You can get away with it on old consoles games that use huge amount of motion blur and other masking techniques where you simply do not see on a telly all kinds of artefacts that checkerboarding and similar generate. But there's a reason they do not go to PC games - it's been tried, and they just looked really bad. Not just blurry but with visible artefacts, especially when most PC players turn off motion blur, set up higher texture details and these games just looking better in the first place on the PC.
I suspect the reason for RT delayed on NVIDIA here is simple - both AMD and NVIDIA require different code for RT (mostly it's down to order of things fed to the driver) to be well optimised for both. Most older games have only code for NVIDIA and they work on AMD very badly and they likely just wrote code for AMD and then worked on NVIDIA's path, which took them a while longer to finish and test before it was released. Since AMD sponsored it, they had priodity.
This is a problem generally with temporal methods, that people seem to praise oh so much, whilst bashing methods like FSR for not having it - they mess up with image, create ghosting and blur it heavily. I'd take Super-sampling any day over TAA but it cost too much performance. With FSR though... Hmm!
That said, even DLSS doesn't use temporal reconstruction or accumulation in a way TAA does it - it uses motion vectors and then temporal feedback to its own up-scaler instead, avoiding the bluring and artefacts that such would bring. Though, ghosting still happens.Basically AMD hardware is less powerful and less able to deal with more divergent rays. The RT GPU performance is more important at this point than the raster performance. If you look at Unreal Engine 5 most of the raster processing is done by the engine in software because no GPU hardware can match the performance. Thus a fast raster proformance on the gpu is less important. If you turn of RT then the performance of the GPU is very important. NVidia cards can provide better DXR performance at higher quality settings and get higher frame times. In every game nvidia is faster. AMD design a card that is faster at current games in raster but left DXR performance low. NVidia gives good raster perfomance and far fastest DXR performance.
This is why DLSS exists, to make DXR games able to run at high resolutions. DLSS 1 used sparial upscaling, an AI netwark and sharpening. This is better than just using sparial upscaling and sharpening for image quality. The media attacked DLSS 1's image quality. We had zoomed in image and pixel by pixel criticism of every deviation from native. So to get the image quality nvidia move to temporal upscaling in DLSS 2.x. Temporal upscaling was the breakthrough for image quality in DLSS 2.x.
AMD is forced to create FSR. FSR uses the same method nvidia rejected for image quality reasons in DLSS 1.x. There is no AI network. You can see the the image is not native, lacks detail and is blurred compared to native. All the problems of DLSS 1. At this point DLSS 2.2 is being called witchcraft its so good at increasing performance and keeping image quality.
There has been no attack on FSR image quaility in the same way as DLSS 1. FSR is blurred and there is less deatil in the image. Youtube posts on a video talking about how good FSR is. This is from a gamer using FSR. They tell the truth.
The massive performance lead in DXR is being ignored. FSR image quality problems are not being addressed. DLSS being ignored for benchmarks results.
Last edited:
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 29,803
Plus the fact Nvidia did it very badly and lazily it seems, AMD's solution seems to be much superior to DLSS 1.xx
Caporegime
That's what the tech press say tbf.