Announcement

Collapse
No announcement yet.

AMD FidelityFX Super Resolution 2.0 Debuts

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #11
    Nice, but how is now compared to Nvidia's thing?

    Comment


    • #12
      Originally posted by aufkrawall View Post
      Remember oiaohm stating here that temporal approaches had no future? Just mentioning it for the lulz...
      If this is can compete with good TAA(U) solutions found in some games, it would be nice progress to have this in a growing number of titles. Unlike FSR 1.0, which was just a step backwards. Really hope AMD can deliver this time.
      I don't know the context here, but temporal scaling has been an absolute pain in the video processing world for over a decade.

      The only half decent implementations I've seen (other than DLSS, which is stupid because it only works in like 2 dozen games) are janky ML networks that take *seconds* to process a frame. Nothing "conventional" seems to work well, no matter how good the mocomp is.


      Originally posted by Danny3 View Post
      Nice, but how is now compared to Nvidia's thing?
      RSR and NIS are both such huge upgrades over classic scaling and close enough to each other (depending on the settings) that the difference is not super important.

      They are both also kind of lacking. AMD in particular needs to merge some of the FSR 2.0 changes into RSR, otherwise they are just holding it back.

      FSR 2.0 and DLSS are completely different animals, even if they seem superficially similar.
      Last edited by brucethemoose; 17 March 2022, 04:04 PM.

      Comment


      • #13
        Oh actually I'm being dense, I forget that FSR2/DLSS get real motion vectors from the engine itself.

        Comment


        • #14
          Originally posted by TNZfr View Post
          Hi,
          Around FSR, I have a little thing to say about RSR :
          • I run a KDE Neon distro on a Ryzen 9 5950x / Radeon rx5700xt with Linux 5.16.15 & Mesa 22.0
          • My desktop session is under wayland.
          • I level up the desktop zoom to 125%. The resolution used by apps is now 3072x1720 instead of 4K.
          This « zoomed » resolution is used by Vulkan and OpenGL games ... and output image are in 4K (monitor OSD 3840x2160@60 Hz). The most impressive are the FPS on my RX5700XT. With « Shadow of the tomb raider » benchmark (Steam linux native), I have a 44 - 45 FPS in 4K resolution ... and 66 FPS with the « zoomed » resolution.

          I'm using this setting since KDE Plasma 5.24.0

          So I wonder about RSR in GNU/Linux distros because what I describe seems to be similar ... maybe a subject for a phoronix test ?

          Setting my desktop to 125% zoom on a Wayland session also caps the resolution used by games to below my monitor's native. I always assumed it was a bug and not how things are supposed to work. Yeah, fps numbers are impressive but the drop in quality is noticeable.

          Comment


          • #15
            Originally posted by brucethemoose View Post
            I don't know the context here, but temporal scaling has been an absolute pain in the video processing world for over a decade.
            Video and 3D rendering are vastly different.

            Originally posted by brucethemoose View Post
            The only half decent implementations I've seen (other than DLSS, which is stupid because it only works in like 2 dozen games) are janky ML networks that take *seconds* to process a frame. Nothing "conventional" seems to work well, no matter how good the mocomp is.
            There is bad TAA(U) in a lot of titles, but some games also get it right (and none uses DL). It's the job of FSR 2.0 to close that gap.

            Originally posted by brucethemoose View Post
            FSR 2.0 and DLSS are completely different animals, even if they seem superficially similar.
            They are at least much closer than DLSS 2.x and FSR 1.0 are.

            Comment


            • #16
              Originally posted by Melcar View Post
              Setting my desktop to 125% zoom on a Wayland session also caps the resolution used by games to below my monitor's native. I always assumed it was a bug and not how things are supposed to work. Yeah, fps numbers are impressive but the drop in quality is noticeable.
              May be ... but I have one of the first 4K monitor : an iiyama prolite b2888 without freesync ... I don't see quality drop. Maybe with a better monitor (HDR10 for example) the drop could be seen.
              Last edited by TNZfr; 18 March 2022, 03:11 AM.

              Comment


              • #17
                Originally posted by brucethemoose View Post
                I don't know the context here, but temporal scaling has been an absolute pain in the video processing world for over a decade.

                The only half decent implementations I've seen (other than DLSS, which is stupid because it only works in like 2 dozen games) are janky ML networks that take *seconds* to process a frame. Nothing "conventional" seems to work well, no matter how good the mocomp is.
                For low-res videos (e.g. 240p MPEG1) I always prefer a slightly blurred video upscaled to 4K over trying to inject artificial "sharpness" with "smart" neural-network upscalers that produce horrendously aliased images which are simply unwatchable.
                Also helps with smoothing over digital video compression artifacts & defects, too.

                Comment


                • #18
                  Originally posted by aufkrawall View Post
                  Video and 3D rendering are vastly different.


                  There is bad TAA(U) in a lot of titles, but some games also get it right (and none uses DL). It's the job of FSR 2.0 to close that gap.


                  They are at least much closer than DLSS 2.x and FSR 1.0 are.
                  Nope they aren't, Dlss works only on 2D final images taking in consideration previous 2D final images. Also there isn't any legit Temporal Upsampling algorithm till today, every one of them introduces artifacts. And when i say artifacts i don't mean image noise like Fsr or ghosting that also other Fx have, i mean particles pop up from another universe and stuff. Till today Temporal Upsampling is worst than lossy graphics.

                  Comment


                  • #19
                    Originally posted by artivision View Post
                    Nope they aren't, Dlss works only on 2D final images taking in consideration previous 2D final images.
                    Guess what FSR 2.0 will do. And of course DLSS 2.x also takes motion vectors into account, you can't do effective TSSAA without them without introducing monstrous ghosting artifacts. You can't re-invent the wheel without making it round, just the tweaking is different...

                    Comment


                    • #20
                      Originally posted by aufkrawall View Post
                      You really can't make that assumption by noticing a slight delay in a video. FSR 2.0 is normal post processing like any other TAA, it shouldn't introduce lag.
                      Have you observed something different? By all means, share your own findings.

                      Comment

                      Working...
                      X