Spatial AA vs. Temporal AA
- Thread starter Gunhead
- Start date
I don't think so. At 60Hz, objects that are too fast-moving to resolve will be out of the screen before you could really pay attention to them anyway. I'd take the added improvement of the look of the motion any day. I don't think the input delay at 60Hz would really be that noticeable compared to 120Hz.Bambers said:
If you are supersampling in the temporal domain, you can poll for input between subframes, effectively reducing input delay.
Chris123234
Regular
For this TAA wont there be an lag time between mouse movement and displayed movement?
If you have to render successive frames to do it, how can it possibly be dont without lag? You cant predict how the person is going to move the mouse so how can the pc renderframes in advance to compensate for the lag?
I see how non-real time thing like movies can do it since they have the frames pre rendered and/or dont need to worry about input lag, but I dont see how games could possibly do it.
If you have to render successive frames to do it, how can it possibly be dont without lag? You cant predict how the person is going to move the mouse so how can the pc renderframes in advance to compensate for the lag?
I see how non-real time thing like movies can do it since they have the frames pre rendered and/or dont need to worry about input lag, but I dont see how games could possibly do it.
Perhaps you could assume that players can't have infinite acceleration rates and so will (approximately) be moving with a constant velocity at least for a few milliseconds.Chris123234 said:
Anyway, these assumptions are often made when you are playing in a networked game since you don't have instaneous communication between nodes.
How many have tried this smartshader?
Try it with a game with very high framerate. And if your framerate goes below 2*vFreq, lower the resolution. I've tried it on Q2, which have plenty of speed. There's seem to be a bug somewhere in smartshader though, since the performance slows down gradually. But it works for a while.
I'd especially like to see comments on what you think about strafeing while looking at a vertical edge. (Especially straight strafeing, not circle strafe around the edge.)
I've always been pro temporal AA, but this gave me a sour taste in my mouth. It seems that I keep tracking objects with my eyes without folowing them with the crosshair, and to a much larger extent than I thought I did.
Temporal AA is great when you always look at the same spot on screen (crosshair), but it do degrade the image quality when you're tracking stuff that move over the screen.
Try it with a game with very high framerate. And if your framerate goes below 2*vFreq, lower the resolution. I've tried it on Q2, which have plenty of speed. There's seem to be a bug somewhere in smartshader though, since the performance slows down gradually. But it works for a while.
I'd especially like to see comments on what you think about strafeing while looking at a vertical edge. (Especially straight strafeing, not circle strafe around the edge.)
I've always been pro temporal AA, but this gave me a sour taste in my mouth. It seems that I keep tracking objects with my eyes without folowing them with the crosshair, and to a much larger extent than I thought I did.
Temporal AA is great when you always look at the same spot on screen (crosshair), but it do degrade the image quality when you're tracking stuff that move over the screen.
Chris123234
Regular
Chalnoth said:
why not? If it has to render future frames to display on the screen now how can there not be lag. Wont it be displaying the screen with information from frames already rendered? Bear with me, Im just not seeing this clearly I guess.
Chris123234
Regular
I thought we were talking like, 5+ frames used for TAA.
Hence the "it depends upon the implementation" statement. There's no reason temporal AA needs to use any specific number of frames blended, or even always blend the same number of frames based upon some setting. One could only render additional frames when the framerate was high enough for those additional frames.Chris123234 said:
I setup scene in 3dmax with boxes moving with various speeds. On each box there was texture with text. Then I rendered animation with motion blur on/off (use full scene MB). Motion blured boxes seemed to move more smoothly but text was hardly readable.I think about per object technique: if difference in position of object between two frames is small, turn motion blur off. Would it be better?
I don't think Temporal AA will be all that noticable even below 60FPS. We are afterall talking about a "slight" shifting of the AA grid which will cause slight colour variances in pixels that are immediately adjacent to one another. Its not like playing a 24FPS DVD, that may/may not be moving half an inch across the screen per frame in high action scenes.
It might be noticable at high contrast and low refresh (black against white, 60FPS or lower) if the convergence of the monitor is off (moire pattern would become pronounced)
This is not AFR, which will introduce lag because at least one image must be prebuffered, and the game runs best in triple buffered mode. Even SLI had a small amount of input lag, but not anywhere near as bad as AFR.
Its a simple AA grid shift in the final pass before it is displayed to the screen. Technically you could probably even put a dongle on the VGA cable to achieve a similar effect. So I am 100 percent positive there is no input lag using this method.
It might be noticable at high contrast and low refresh (black against white, 60FPS or lower) if the convergence of the monitor is off (moire pattern would become pronounced)
This is not AFR, which will introduce lag because at least one image must be prebuffered, and the game runs best in triple buffered mode. Even SLI had a small amount of input lag, but not anywhere near as bad as AFR.
Its a simple AA grid shift in the final pass before it is displayed to the screen. Technically you could probably even put a dongle on the VGA cable to achieve a similar effect. So I am 100 percent positive there is no input lag using this method.
Thanks to everybody for definitions, clarifications, examples, and links.
TAA got explained very thoroughly. I really love these discussions -- they are quite the forté of B3D. (In addition to the excellent articles, of course.)
And, incidentally, I guess it's also pretty clear now that ATI Marketing has slightly erred on the BS side of things with their "Temporal FSAA"
TAA got explained very thoroughly. I really love these discussions -- they are quite the forté of B3D. (In addition to the excellent articles, of course.)
And, incidentally, I guess it's also pretty clear now that ATI Marketing has slightly erred on the BS side of things with their "Temporal FSAA"
Bjorn said:Gunhead said:And, incidentally, I guess it's also pretty clear now that ATI Marketing has slightly erred on the BS side of things with their "Temporal FSAA"
Maybe i've missed something but has Ati directly marketed TAA in any way ?
R420 launch I believe they had info on it if not they gave it to the reviewers.
Bjorn said:
From Ati.com -> products, gamer -> X800 -> specifications:
SMOOTHVISIONâ„¢
* 2x/4x/6x Anti-Aliasing modes
o Sparse multi-sample algorithm with gamma correction, programmable sample patterns, and centroid sampling
o Lossless Color Compression (up to6:1)at all resolutions, including widescreen HDTV resolutions
o Temporal Anti-Aliasing
* 2x/4x/8x/16x Anisotropic Filtering modes [...]
IMO you are directly marketing if you advertise it on the company website
Gunhead said:IMO you are directly marketing if you advertise it on the company website
That's of course true. What i meant was that they haven't really gone out and said that TAA is best thing since sliced bread so accusing them for marketing BS doesn't really seem fair. TAA has many drawbacks but it might still be a good option for some games.
