Sorry for poor mic quality at times. I boosted my mic so I could be heard better but I did it too much so it sounds off at times, I will fix it for next videos. Also when looking at shadow detail in the video, I should've zoomed in even more because it's very hard to see what I'm talking about after UA-cam decompression + watching on my phone
@@reimner I didn't have the DLSS screenshots then, I got those afterwards. But please note I'm not talking about the shadow issue from shadows getting lighter and less detailed, I'm talking about the issue of FSR 2.0 having random lines through the shadows and looking more blocky sometimes than FSR 1.0 and native. However you can't see this in the video as I didn't zoom in enough to compensate for YT decompression
you could've used more time on analysing FSR 2.0 in motion on deathloop. It is quite impressive, but i suggest looking at the thin lines/wires, and looking at the textures when two objects meet. Hopefully, there will be a future video on this. Anyways, thx for this analysis!
Yes, but their are two issues with that video. 1. It's a video not a screenshot, therefore it's even more compressed and the video itself was running at quite a low bitrate as well 2. The videos motion was very slow, deliberately so, which will reduce ghosting and artifacts which any we've seen could've been from the low bitrate anyways. So although I agree with you and wanted to do that, we need better marketing material from AMD or 3rd party videos to analyze because as is this one wasnt good enough to work with. Thank you for your comment
You should've focused more on quality than you did imo. Myself, and many others will probably stick to the better quality settings (at least balanced and above). So, the quality FSR 2.0 compared to native and quality FSR 1.0 would be a better comparison. Also it is odd to focus on the difference between native and performance mode, when performance mode is more about, well performance. Maybe a second video going in depth with the quality modes would be nice. I personally have downloaded the images, and a lot of the problems FSR 2.0 has aren't as noticeable in the quality mode, and impossible for me to see when not zoomed in (except for very slightly brighter shadows, which again quality mode does much better with).
I did look at quality mode here, but there were simply more performance screenshots and the lower the resolution the image has to reconstruct from the better we can find the flaws with the construction technique and how it works. I think most will use quality, but in games with ray tracing performance mode will be more popular probably so people will use both a lot. These are my reasons for why I compared the performance preset, also I did compare it to FSR 1.0 in the video too luckily, I had FSR 1.0 on preset higher to make it fair as well.
"Here at AMD we have to fullfill our commitment to improving the gaming scene by modding fucking lanczos, the 20+ year old algorithm and rebranding it, then threating it like it's as good as a temporal ai solution. we also fooled the majority of pc gamers that cant see anything because of their hate towards nvidia dlss not supporting the gtx 1000 series even tho it doesnt have the cores to run it at more than 2 fps by using our cool sharpening, so that they thought the image quality was as good as native. then, we started modding temporal anti aliasing upsampling, a technology that has been available for 3 years in the majority of games, and now the pc gaming community thinks we killed dlss! by open sourcing all of our, sorry, other people's code, we were able to make people think we are good. we are on our way to develop the state-of-the-art, next gen anti aliasing technique that is literally smaa 2tx but with fidelityfx in the name." disclaimer: i only hate AMD for the software, not the hardware. i have a ryzen 7 3700x and it's pretty fucking good. in fact, all of AMD problems are software side (eg. poor drivers) i'm willing to eat back all of this vomit if fsr 2.0 is actually impressive and something more than modded TAAU
I work with FSR 1.0, and TAAU. TAAU was built into our game engine at Ubisoft, and from looking at the source code of FSR 1.0 it's more advanced than Lanczos and it shows in results. Its edge detection is superior and their RCAS boosts details, resulting in superior image quality. As for FSR 2.0 I need to get my hands on it first but from looking at the released images, typically TAAU doesn't cause the image to look like the native resolution of your monitor it very obviously doesn't do that when it warps things to smooth it out a bit once you get to low resolutions, FSR 2.0 is doing some superior reconstruction. But what I'd like to point out first is that DLSS relies on Temporal Upscaling and most of the results from the DLSS image are actually from TAAU much like FSR, the AI is only responsible for picking from a list of TAAU results for the best one, the work it does is minimal it just gives it that little extra edge. Former versions of DLSS used AI a lot more than 2.0 did, DLSS 2.0 realized AI upscaling isn't entirely the future, traditional temporal upscaling paired with AI is. If you watched AMD's GDC conference on how FSR 2 works you'd see it's very different internally from TAAU, and more advanced. It's a long video but if you have time I recommend it.
Sorry for poor mic quality at times. I boosted my mic so I could be heard better but I did it too much so it sounds off at times, I will fix it for next videos.
Also when looking at shadow detail in the video, I should've zoomed in even more because it's very hard to see what I'm talking about after UA-cam decompression + watching on my phone
And also should have compared them with dlss cuz it was also doing the same thing with shadows for some reason. Maybe deathloop is broken idk
@@reimner I didn't have the DLSS screenshots then, I got those afterwards.
But please note I'm not talking about the shadow issue from shadows getting lighter and less detailed, I'm talking about the issue of FSR 2.0 having random lines through the shadows and looking more blocky sometimes than FSR 1.0 and native. However you can't see this in the video as I didn't zoom in enough to compensate for YT decompression
One of the best channel i came across
you could've used more time on analysing FSR 2.0 in motion on deathloop. It is quite impressive, but i suggest looking at the thin lines/wires, and looking at the textures when two objects meet. Hopefully, there will be a future video on this. Anyways, thx for this analysis!
Yes, but their are two issues with that video.
1. It's a video not a screenshot, therefore it's even more compressed and the video itself was running at quite a low bitrate as well
2. The videos motion was very slow, deliberately so, which will reduce ghosting and artifacts which any we've seen could've been from the low bitrate anyways. So although I agree with you and wanted to do that, we need better marketing material from AMD or 3rd party videos to analyze because as is this one wasnt good enough to work with. Thank you for your comment
You should've focused more on quality than you did imo. Myself, and many others will probably stick to the better quality settings (at least balanced and above). So, the quality FSR 2.0 compared to native and quality FSR 1.0 would be a better comparison. Also it is odd to focus on the difference between native and performance mode, when performance mode is more about, well performance. Maybe a second video going in depth with the quality modes would be nice. I personally have downloaded the images, and a lot of the problems FSR 2.0 has aren't as noticeable in the quality mode, and impossible for me to see when not zoomed in (except for very slightly brighter shadows, which again quality mode does much better with).
I did look at quality mode here, but there were simply more performance screenshots and the lower the resolution the image has to reconstruct from the better we can find the flaws with the construction technique and how it works.
I think most will use quality, but in games with ray tracing performance mode will be more popular probably so people will use both a lot. These are my reasons for why I compared the performance preset, also I did compare it to FSR 1.0 in the video too luckily, I had FSR 1.0 on preset higher to make it fair as well.
Nice
Add fsr2.0 on lossless scaling
"Here at AMD we have to fullfill our commitment to improving the gaming scene by modding fucking lanczos, the 20+ year old algorithm and rebranding it, then threating it like it's as good as a temporal ai solution. we also fooled the majority of pc gamers that cant see anything because of their hate towards nvidia dlss not supporting the gtx 1000 series even tho it doesnt have the cores to run it at more than 2 fps by using our cool sharpening, so that they thought the image quality was as good as native. then, we started modding temporal anti aliasing upsampling, a technology that has been available for 3 years in the majority of games, and now the pc gaming community thinks we killed dlss! by open sourcing all of our, sorry, other people's code, we were able to make people think we are good. we are on our way to develop the state-of-the-art, next gen anti aliasing technique that is literally smaa 2tx but with fidelityfx in the name."
disclaimer: i only hate AMD for the software, not the hardware. i have a ryzen 7 3700x and it's pretty fucking good. in fact, all of AMD problems are software side (eg. poor drivers)
i'm willing to eat back all of this vomit if fsr 2.0 is actually impressive and something more than modded TAAU
I work with FSR 1.0, and TAAU. TAAU was built into our game engine at Ubisoft, and from looking at the source code of FSR 1.0 it's more advanced than Lanczos and it shows in results. Its edge detection is superior and their RCAS boosts details, resulting in superior image quality.
As for FSR 2.0 I need to get my hands on it first but from looking at the released images, typically TAAU doesn't cause the image to look like the native resolution of your monitor it very obviously doesn't do that when it warps things to smooth it out a bit once you get to low resolutions, FSR 2.0 is doing some superior reconstruction.
But what I'd like to point out first is that DLSS relies on Temporal Upscaling and most of the results from the DLSS image are actually from TAAU much like FSR, the AI is only responsible for picking from a list of TAAU results for the best one, the work it does is minimal it just gives it that little extra edge. Former versions of DLSS used AI a lot more than 2.0 did, DLSS 2.0 realized AI upscaling isn't entirely the future, traditional temporal upscaling paired with AI is. If you watched AMD's GDC conference on how FSR 2 works you'd see it's very different internally from TAAU, and more advanced. It's a long video but if you have time I recommend it.
This is complete nonsense.. i wasted 5min on "this"
Well I'm glad you only wasted 5mins on the video then and not 13 if you feel like it was a waste of time. Thanks for giving it a chance!