NVIDIA G-SYNC: Variable Refresh Rate Monitors!

Posted Oct 19, 2013 by

One small step closer to tomorrow’s Holodeck:Β NVIDIA G-SYNC! It is a technique for refreshing computer monitors at variable intervals (up to a certain limit).Β Instead of refreshing monitors at exact intervals, the monitor is refreshed when the GPU finishes generating a frame! [Image: AnandTech liveblog of launch event]

Variable refresh rates combines advantages of VSYNC ON (eliminate tearing) with the advantages of VSYNC OFF (low input lag), while virtually eliminating stutters (no mismatch between frame rate and refresh rate).

It is confirmed that this, alone, won’t eliminate motion blur as completely as strobe backlights (e.g. LightBoost), but this is a great step towards eliminating discrete refresh rates (which creates motion blur even at 144Hz. See photos of 60Hz vs 120Hz vs LightBoost), especially as the 144Hz limit is raised in the future, while using nVidia G-SYNC. Hopefully strobe backlight technologies can be combined with G-SYNC in the future — and hopefully already in some upcoming models.

EDIT:
– All G-SYNC monitors include an official strobe backlight mode, better than LightBoost!
– Mark Rejhon has quickly come up with a new method of dynamically blending PWM-free backlight at lower framerates to strobing at higher framerates; see addendum to Electronics Hacking: Creating a Strobe Backlight. This allows combining LightBoost + G-SYNC without creating flicker during lower framerates!

 

About Mark Rejhon

Also known as Chief Blur Buster. Founder of Blur Busters. Inventor of TestUFO. Read more about him on the About Mark page.

39 Comments For “NVIDIA G-SYNC: Variable Refresh Rate Monitors!”

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Sort by:   newest | oldest | most liked
snakeeyes
Member
snakeeyes

hey i’ve got a question, can you use both modes G synic and strobe modes at the same time ? cause that’s the only reasen that would make me dump amd’s mantle and true audio for a cheaper price.

Like0Vote Down 
10 years 6 months ago
Mark Rejhon
Author
Mark Rejhon

Not yet, but it’s theoretically possible. I’m hoping NVIDIA is working on it.
I have a flicker-free variable rate strobe backlight algorithm, and offered it for free to NVIDIA:
Electronics Hacking: Creating a Strobe Backlight
(I am only asking for credit from NVIDIA)

Like0Vote Down 
10 years 6 months ago
snakeeyes
Member
snakeeyes

you really seem to have worked very hard for this i’ve been on this site many times and you never rested so i hope you get all the credit cause you deserve it as a hard worker πŸ™‚

Like0Vote Down 
10 years 6 months ago
Mark Rejhon
Author
Mark Rejhon

Thanks! Blur Busters is adding extra stuff rapidly for you all, and we’ve added the new G-SYNC menu at the top.

There are several additional big sections being added (other than LightBoost and G-SYNC). Keep tuned to Blur Busters over the next few weeks!

Like0Vote Down 
10 years 6 months ago
snakeeyes
Member
snakeeyes

and get a 4k monitor ofc instead of all this

Like0Vote Down 
10 years 6 months ago
blargg
Member
blargg

kirkh, couldn’t it just lengthen the next strobe? If the variation in frame duration was too great, it could insert an extra frame, rather than have brightness fall a lot, then raise a lot during the next few frames to compensate. This would only occur when the program was varying the rate drastically.

Like0Vote Down 
10 years 6 months ago
Mark Rejhon
Author
Mark Rejhon

This is definitely a workable idea too.
(I’ve run the thought experiment, and it actually works; using the previous frame’s information to generate the strobe curve for the current frame)

Like0Vote Down 
10 years 6 months ago
kirkh
Member
kirkh

Wouldn’t you need to know the duration of the current frame to adjust strobing appropriately? And if so doesn’t that imply at least one frame of latency? I think your solution works, curious to see what Nvidia has cooked up, maybe its the same thing.

Like0Vote Down 
10 years 6 months ago
Mark Rejhon
Author
Mark Rejhon

Two solutions.

#1: Not a full frame of latency. You only need a partial-frame FIFO buffer roughly equalling to the longest planned strobe length. Pixel in from display cable, pixel out to display. Buffer only delays refreshes by, say, 1 millisecond. You scan-out the current refresh (in darkness) anyway delayed only by one or two extra milliseconds, and then begin your strobe the moment the next frame arrives on the display cable, strobe for a millisecond, and then turn off, begin refreshing the next frame in darkness. Obviously, this complicates the strobe curve calculations a bit, but you can limit your strobe-flattening and disable your strobing below a specific framerate. It would only add an extra 1-2ms of input latency, if your longest planned strobe curve was 1ms.

#2: No added latency (beyond average half the frame scan-out time; LightBoost adds an average ~4ms of input latency). For strobed G-Sync, you can use previous frame’s frametime information to decide on a strobe length for the current frame. And since you stop strobing below a certain refresh rate, there appears to be no side effects from doing this, as long as you don’t let the strobe length variations interfere with beginning of the next frame refresh (by, possibly, ensuring a large enough blanking interval margin between refreshes). It’s easiest if you choose very short strobe lengths, 0.5ms sharp wave at 120Hz, flatter 0.75ms wave at 90Hz, faint 1ms wave at 60Hz, and cutting off to PWM-free below that. So you just guarantee an extra 1ms of safety margin into your blanking intervals. This is already being done with LightBoost (which goes down from 144Hz to 120Hz when you enable strobing), so G-Sync might cap out at 120fps instead of 144fps if you enable strobing. Nothing new here, and it would still have less input lag than LightBoost (due to G-Sync’s refresh-on-demand ability)

Like0Vote Down 
10 years 6 months ago
Mark Rejhon
Author
Mark Rejhon

To accomplish LightBoost + G-Sync, you need very precise control of strobe backlight strobe lengths during variable-rate strobing, to maintain constant brightness, and you need to maintain a minimum framerate to prevent low framerates from creating visible flicker (a downside of LightBoost + G-Sync during lower framerates is doable without flicker.

One solution I believe that is being considered is a gradual transition from strobing (at higher refresh rates) blending into PWM-free mode (at lower refresh rates) in a smooth dynamic control of the backlight strobing curve.

Square-wave strobing at 120Hz blends into softer CRT-style gentle strbing curve at 85Hz, blends into a flat PWM-free steady backlight at 60Hz and lower. All maintaining at exactly the same perceived brightness. My engineering mind tells me G-Sync + LightBoost is possible with creative dynamic control of the LightBoost strobe curve, to prevent flicker at lower frame rates.

Such dynamic strobe-blending-to-nonstrobed prevents flicker at lower framerates, while giving you full LightBoost benefits at higher framerates. Using completely seamless strobe-length/strobe-curve manipulation from a flat line at lower framerates, all the way to square-wave strobing at high framerates beyond flicker fusion threshold.

John Carmack’s briliant mind might have already imagined this; but I’ve tweeted this comment to him (to find out if he already came up with this idea) as he pays attention to Blur Busters from time to time.

Also from a John Carmack twitter Reply, the VG248QE G-Sync upgrade also has improved LightBoost as a secondary option, even if not yet able to be simultaneously enabled with G-Sync:

John Carmack ‏@ID_AA_Carmack 10h
@GuerrillaDawg the didn’t talk about it, but this includes an improved lightboost driver, but it is currently a choice — gsync or flashed.

I believe I may have invented a method of blending PWM-free and strobing:

Example:
10fps@10Hz β€” PWM-free backlight
30fps@30Hz β€” PWM-free backlight
45fps@45Hz β€” PWM-free backlight
60fps@60Hz β€” Minor brightness undulations (bright / dim)
80fps@80Hz β€” Sharper brightness undulations (very bright / very dim)
100fps@100Hz β€” Resembles rounded-square-wave (fullbright / fulloff)
120fps@120Hz and up β€” Nearly square-wave strobing like original LightBoost

This would be a dynamically variable continuum all the way in bewteen too, much like automobile CVT instead of discrete gears in automobile transmissions. You avoid flicker at lower frame rates, and you get full strobing benefits at higher frame rates.

Simpler algorithsm variations are also possible (e.g. keeping a square wave, and using only pulsewidth / pulseheight manipulation to achieve the blending effect, but without curve-softening). This is included as part of my general idea of blending from PWM-free at lower refresh rates, to strobing at higher refresh rates.

I have updated Electronics Hacking: Creating a Strobe Backlight to include this PWM-free-blending-to-strobing idea that is being offered here.

That said, in the event that neither nVidia nor John Carmack thought of this creative strobe-blending-to-PWM-free algorithm to prevent strobe flicker at lower framerates, please give either Mark Rejhon and/or Blur Busters a call-out on your website. Although I normally prefer my ideas not get patented by others without consulting me first, if my idea suggestion is patentable, please give appropriate due written credit (e.g. To my name, or to Blur Busters) in your patent application. That’s the only credit I ask for this free idea (no charge from me for this specific idea, provided credit is given). We, at Blur Busters, totally support innovative display progress such as this! Free idea being offered; no financial compensation is being asked for.

Thanks,
Mark Rejhon
Chief Blur Buster
http://www.blurbusters.com/about/mark
October 19th, 2013

Like0Vote Down 
10 years 6 months ago
wpDiscuz

Recent Content

Blur Buster Forums

Blur Busters Forums

TestUFO Motion Tests

TestUFO Motion Tests