Page 1 of 1

Just remembered that in 2013 I predicted G-Sync by chance...

Posted: 16 Mar 2017, 22:07
by RealNC
I was having a discussion on the nvidia forums about vsync, and said that "it's about time someone invents a variable refresh rate technology." [sic]

This was before g-sync was even announced for the first time. Lo and behold, half a year later g-sync was announced... However, I was wrong when I claimed that "I don't see that happening any time soon." :lol:

https://forums.geforce.com/default/topi ... 4/#3790524

Re: Just remembered that in 2013 I predicted G-Sync by chanc

Posted: 17 Mar 2017, 08:33
by Chief Blur Buster
Variable refresh rates for raster-scanned displays was not something I predicted. Not at all.

The closest I did to predicting GSYNC/FreeSync -- was to predict it was possible to use graphics-card-side "video genlocking" by adding/removing scanlines to the blanking interval, to keep a graphics card exactly in sync with a video signal.....back in year 2004.

However, this website was formerly http://www.scanningbacklight.com -- I did predict scanning/strobe backlights could allow LCD to finally do CRT-clarity motion back 2011-2012. LightBoost was discovered which changed Blur Buster's direction. I was able to recognize that GtG was potentially only a small limiting factor on 3D-capable LCDs due to the way they cleaned up their refresh cycles (during the black interval) nearly completely before the next refresh cycle.

Back at the time, so few people were convinced that LCD could now achieve the feats of CRT-clarity motion -- at least when the strobe backlight is done properly.

Re: Just remembered that in 2013 I predicted G-Sync by chanc

Posted: 17 Mar 2017, 12:41
by RLBURNSIDE
Enthusiasts' ideas are often incorporated (without credit let alone renumeration) into companies' product lines, surreptitiously. Although I do think VRR has been supported by eDP for laptops for many years, only for saving power instead of gaming.

VRR both makes sense and doesn't.

I write games VR now? And guess what, you need a stable 90+ fps, with low persistence for a smooth experience. In a headset you would notice saccaded rendering rhythm such as Freesync. It came up many times, the question :"why no VRR in Rift?"

In VR, they use frame compositors to "just-in-time" reproject or synthesize a new frame to maintain solid 90fps or 120fps (Morpheus renders at 60 fps all the time AFAIK and so this reprojection is always on, which also makes perfect sense given the GPU requirements).

What this begs is: Why don't 30 fps games use reprojection to hit 60fps, similar to morpheus 60 -> 120, and Rift/ Vive's 45 -> 90. Remember, if your frame takes over 11ms (1/90), your framerate is cut in half to 45 for that frame, resulting in stutter. Oscillating between 45 and 90, randomly, is extremely irritating, and even real frames vs reprojected ones don't look the same, obviously, so what you really want to do is render at a stable 45 fps + reproject to 90, or render at a stable 90 fps (and lose image quality probably, due to tradeoffs). So reprojection is like a safety net, and not something that is meant to be used, if targetting 90 fps your engine needs to try to do that on its own. And if you can't hit stable 90fps, you should instead pack as much detail into 45 fps frames as possible, and use reprojection on all the time. This will give a consistent look. This is the reason why Morpheus games are rendered at 60 fps instead of 120. It's hard enough for PS4 games to hit 60 most of the time, especially in stereo.

What I'm saying is: VRR is a neat trick, but ideally, you don't want variable frame duration, but reprojected frames to reduce temporal aliasing and input lag variance. If some frames naturally take 20 ms and others take 13ms, then you're going all over the place between 30 and 60 fps (typical gaming scenario pre-optimisation), and that variance will show itself to your eye. I like G-sync and so on, but I really think framerates should be unbounded, and the only frames the player ever sees should be of the reprojected and temporally re-sampled variety.

There is tech out there now for phase-space frame interpolation, acceleration and deceleration to do this in screen space, although in VR it's harder because screen space techniques don't work all that well (eye misconvergence). So VR reprojection needs the depth buffer. Which is fine because the frame compositor has the entire swap chain + depth to play with.

I really think the future of smooth gaming lies in innovations from VR tech being brought back into traditional TV / monitor 2D games. Time will tell.

Re: Just remembered that in 2013 I predicted G-Sync by chanc

Posted: 17 Mar 2017, 22:23
by Chief Blur Buster
RLBURNSIDE wrote:Enthusiasts' ideas are often incorporated (without credit let alone renumeration) into companies' product lines, surreptitiously. Although I do think VRR has been supported by eDP for laptops for many years, only for saving power instead of gaming.
And a 20 year old multisync CRT works with FreeSync VRR (via ToastyX + HDMI->VGA adaptor). At least those CRTs that do 'noisy mode change' -- the kind that doesn't do an intentional electronic blankout during refresh rate changes. Vector CRT's were always VRR by design, but little did we know that raster multisync CRTs actually works with FreeSync tricked to run over a VGA output (about 3 tubes have been successfully tested by various individuals on the Net)...

Re: Just remembered that in 2013 I predicted G-Sync by chanc

Posted: 20 Mar 2017, 17:27
by RealNC
RLBURNSIDE wrote:What I'm saying is: VRR is a neat trick, but ideally, you don't want variable frame duration, but reprojected frames to reduce temporal aliasing and input lag variance.
From how I understand reprojection, it really only works well for camera panning. Which is why it suits VR well. Outside of VR, it looks like it's only suitable for first person games. And then only for games that have a centered viewport. Some games don't even have that (No Man's Sky, for example.)

Getting that to work for non-panning motion (like animation or motion where geometry changes) seems very difficult.