TECHNICAL ANALYSIS: “So what refresh rate do I need?”

Posted Feb 23, 2014 by

A very interesting analysis for geeks. It links to many articles (including by Valve Software) and scientific papers, as well as touches upon why ultrahigh frame rates (e.g. 1000fps at 1000Hz) is the only way to achieve flicker-free low persistence without strobing, phosphor, black frames or similar light modulation.

Read Technical Analysis: “So what refresh rate do I need?”
(See Page 1, Page 2, Page 3 of Blur Busters Forum Area 51 thread)

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.

7 Comments For “TECHNICAL ANALYSIS: “So what refresh rate do I need?””

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

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

There is a hope for high fidelity graphics and 1000+ FPS in the future:
– foveated rendering: (EA DICE) http://www.dailymotion.com/video/x1ehtw4_johan-andersson-foveated-rendering_tech
– some modern precise Anti-Aliasing methods use motion vectors
– Killzone on PS4 uses temporal reprojection: http://www.killzone.com/en_GB/blog/news/2014-03-06_regarding-killzone-shadow-fall-and-1080p.html
– frame interpolation in games can use true data about motion and depth instead of just guessing like TV with video (or 2D->3D automatic converters); Oculus do a research with Time warping (http://www.altdevblogaday.com/2013/02/22/latency-mitigation-strategies/ , http://www.roadtovr.com/gdc-2014-oculus-session-working-latest-oculus-rift-hardware-software-live-blog-11am-pst/) and with a combination of both: http://youtu.be/Uooh0Y9fC_M?t=1h2m25s

Like0Vote Down 
10 years 28 days ago
Mark Rejhon
Author
Mark Rejhon

Good reply with good links! Thanks for posting this, konti!

Like0Vote Down 
10 years 27 days ago
SS4
Member
SS4

Its true that if things goes at the same pace it is going right now it won’t happen soon.
But we never know when someone will come out with something that will revolutionize display in a way we can’t imagine so we can at least hope.
Evolution as well as technology has been known to have made abnormal growth at time.

Like0Vote Down 
10 years 1 month ago
Arael
Member
Arael

1000fps at 1000Hz is probably beyond our lifetime, especially when we observe constant slowdown of computer hardware performance gains. If top GPUs till 2020 will achieve ~3x more performance necessary for 2160p gaming at the same quality as we have now in 1080p, then it still will be something – compare it to 1000fps expectations. Moore’s Law is dead, so are our dreams.

Like0Vote Down 
10 years 1 month ago
Mark Rejhon
Author
Mark Rejhon

You never know. There’s been 1000fps@1000Hz monochrome DLP displays in research laboratories. Also, there’s already a commercially available scientific display: Native 500fps@500Hz display at http://www.vpixx.com

Like1Vote Down 
10 years 1 month ago
Arael
Member
Arael

I do believe OLED should enable industry to create 1000Hz displays by the sole nature of LED that OLED shares, which is ability to operate at high frequencies. BUT I see no chances for 1000fps content, maybe except stuff like OS desktop (and even if so it’s whopping 5.8GB/s to compose and transmit just for 1080p! …and full capacity of single layer BluRay disk thrown at display every second in case of 2160p!). Even if miracle will happen and we’ll reach ~17x performance of GPUs to have 1000fps in current games then 17x more demanding games will appear almost instantly, so 1000fps will stay realistic only for very old content (like Quake Live currently). And we have also other milestones to achieve in case of displays, like hiDPI (finally!) and trueHDR (who knows when, but it’s a must to mimic what our eyes can see and we’re crippled in terms of computer graphics without it). I wish for myself and for everyone blur-free displays, but I don’t see “1000Hz/1000fps” as a realistic approach. Currently I’m looking forward for depth-in OLED TV tests, because I’m “sample-and-hold” theory sceptic (to a certain degree – even if display without transition stage, that is with close to instant response will suffer because of it, then high-frequency black frame insertion should do the trick without mess of extra backstrobe). I hope there is some solution that can enable us to achieve perfect blur-free motion in case of display reproducing variable framerate (which seems to be another great and essential but long-postponed accomplishment IMO). Well, we’ll see – if we aren’t talking about great performance gains out of silicon, but about display technology improvements, then there are still some foreseeable opportunities.

Like1Vote Down 
10 years 1 month ago
Mark Rejhon
Author
Mark Rejhon

Yes, OLED will be very helpful. We’re looking forward to good OLEDs arriving.

However, fact check time. The word “persistence” is essentially a synonym for the “sample-and-hold” effect, though persistence implies it can be a curve rather than squarewave (sample and hold), but persistence can also be squarewave (e.g. turning on/off OLED, or turning on/off backlight).

Also, OLED has a motion blur problem if it’s not strobed:

HDGuru OLED TV Review of LG 55″:
http://hdguru.com/lg-55ea9800-oled-hdtv-reviewed/
“Another test that seemed to back up this hypothesis is our motion resolution test, which provided the same result as 60 Hz LED LCD (a disappointing 320 lines of resolution per picture height).”

Japanese scientist paper:
http://www.journalofvision.org/content/13/7/6.full
See figure 20 with a chart where OLED is worse than plasma (plasma is closer to CRT)

BlurBusters:
https://blurbusters.com/faq/oled-motion-blur/
This scientifically explains why OLED has more motion blur

Gamers:
http://community.eu.playstation.com/t5/Support-PS-Vita/new-OLED-motion-blur/td-p/15180223
Players of PS Vita are frequently noticing slight amount of motion blur.

Currently I’m looking forward for depth-in OLED TV tests

Many parties including CNet, HDTVTest, us, and others, have confirmed OLED motion blur. See http://reviews.cnet.com/8301-33199_7-57612490-221/led-lcd-vs-oled-vs-plasma/ …. It’s confirmed. Non-flickering OLED has tons of motion blur, just like LCD.

John Carmack and Oculus confirms that OLED creates motion blur if you don’t strobe it. That’s why the Oculus Crystal Cove prototype use strobing (there’s no difference between “strobing” and “black frame insertion” when it comes to OLED). The Oculus Crystal Cove prototype is an impulse-driven OLED, which flashes once per refresh.

because I’m “sample-and-hold” theory sceptic

Silly guy. It’s not a theory. Persistence (essentially a synonym for “sample-and-hold”) is a huge problem where it is critical such as virtual reality. Also, another word for “sample-and-hold” is called “persistence”, which represents the duration of a visible frame. There are a lot of different terminologies used in the industry, but they all actually boil down to the same thing.

Also, you should also be aware of the animation http://www.testufo.com/eyetracking which is a “see-for-yourself” proof of the motion blur problem. Also you should be aware of the TestUFO Panning Map Test — street name labels will be unreadable on any display that has more than about 2-3ms of persistence. In addition, the proof of persistence blur (sample-and-hold blur) is easily witnessed by seeing the checkerboard effect in this animation on a modern fast-responding non-impulse-driven display (e.g. recent LCD). Make sure you’re using a stutter-free web browser listed at http://www.testufo.com/browser.html …. Even DLP projectors and most OLED screens display the checkerboard effect, showing their high persistence effect (sample hand hold effect). Try photographing the checkerboard illusion with a camera — you can’t if you hold the camera stationary — you can only capture the checkerboard successfully if you pan the camera along the test pattern. (Excellent demonstration of panning camera motion blur equivalence of eye tracking motion blur). Just try it yourself. 🙂 Low persistence is like a faster camera shutter or a faster camera flash (one flash per refresh), as already proven by adjustable-persistence monitors (if you own one of them). Shorter strobe flash is equivalent to fast shutter. 2ms persistence = 1/500sec = same as 1/500sec shutter speed during same speed camera lens panning (as a camera equivalent to eye tracking). Photos taken with 1/500sec shutter has less motion blur than photos taken with 1/60sec shutter.

then high-frequency black frame insertion should do the trick without mess of extra backstrobe

From the OLED perspective, there is no difference between strobing and black frame insertion. It’s the same kind of blackness to the human vision. The only critically important criterion for zero motion blur is one flash of the frame per refresh. Otherwise, you get PWM artifacts similiar to the 30fps@60Hz CRT double-edge effect, or the 60fps@120Hz LightBoost double-edge effect, or 60fps@180Hz PWM dimming triple-edge effect. Just view http://www.testufo.com on a CRT and you’ll instantly see that half-framerate creates double-image effects, and quarter-framerate creates quad-image effect. Or if you own a PWM dimmed monitor, just view http://www.testufo.com/ghosting at Brightness=0% and you will understand why it’s bad — e.g. 360Hz PWM on a 60Hz LCD creates a six-image effect rather than a comfortably continuous motion blur. For the zero motion blur effect without artifacts, you must have only one flash per frame. CRT flicker, LightBoost flicker, OLED flicker, black frame insertion, whatever, doesn’t matter much how the blackness happens between frames to reduce persistence, except that softening the transition between on/off such as phosphor decay effects or slow GtG will be accumulating/increasing the motion blur above-and-beyond the formula).

I hope there is some solution that can enable us to achieve perfect blur-free motion

Motion blur is directly proportional to persistence. This effect is also easily witnessed in adjustable-persistence monitors (e.g. BENQ Z-series with new firmware that has strobe lengths adjustable from 0.5ms to 4.0ms).

At the end of the day — It doesn’t matter HOW you shorten persistence:
– Shorten persistence via extra black period between refreshes.
– Shorten persistence via black frame insertion.
– Shorten persistence via turning off OLED between refreshes.
– Shorten persistence via turning off backlight on LCD betwee refreshes.
– Shorten persistence via phosphor decay (CRT phosphor that goes dark between refreshes)
– Shorten persistence via intermediate frames via motion interpolation
– Shorten persistence via real frames
– Shorter persistence can be shorter decay phosphor

1ms of persistence translates to a minimum of 1 pixel of motion blur during 1000 pixels/second motion; this formula is quite accurate for square-wave transitions (e.g. strobe backlights or OLED modulation). You can halve motion blur (halve persistence) via adding a symmetrical black frame (50%-50%) or by doubling the frame rate (via interpolation or via native frame rate). You can reduce 75% of motion blur (quarter persistence) via being dark 75% of the time, or you can add three intermediate frames (via interpolation or via native frame rate). Doesn’t matter. The motion blur scales with the duty cycle if you’re keeping frame rate constant and adjusting the amount of blackness between refreshes (e.g. 25%:75% bright:dark via strobing, via black frame insertion). The formula holds regardless of how you shorten persistence. The bottom line is once your transitions become instantaneous and you still have lots of motion blur, you still need to reduce persistence by shortening frame visibility time. Persistence is not the same thing as GtG. Also, it is well known by looking at motion interpolation, that motion interpolation reduces LCD motion blur. Double framerate halves motion blur of video, and quadruple framerate creates one-quarter motion blur of video (during moments that interpolation is working well).

Also, good “see-for-yourself” example of how motion blur scales with duty cycle:
(View this on a 120Hz flickerfree monitor such as LCD, so it runs 30fps with varying levels of black frame insertion)
http://www.testufo.com/#test=blackframes&count=4&equalizer=1
– 100% bright is full motion blur
– 75%:25% bright:dark is 25% less motion blur at 30fps @ 120Hz
– 50%:50% bright:dark is 50% less motion blur at 30fps @ 120Hz
– 25%:75% bright:dark is 75% less motion blur at 30fps @ 120Hz

If only have a 60Hz flickerfree monitor such as an LCD or iPad, it will flicker too much, so try http://www.testufo.com/#test=blackframes&count=3&equalizer=1 instead if you have a 60Hz display:
– 100% bright is full motion blur
– 66%:33% bright:dark is 33% less motion blur at 20fps @ 60Hz
– 50%:50% bright:dark is 50% less motion blur at 20fps @ 60Hz
– 33%:66% bright:dark is 66% less motion blur at 20fps @ 60Hz

Obviously, you need to run the educational TestUFO animations in a stutter-free browser (see http://www.testufo.com/browser.html for system requirements). Now that your jaw has dropped that persistence is no longer a theory, you can pick up the jaw off the floor, and re-read several of the Blur Busters with your newfound education of motion blur behaviors demonstrated by many of the links I’ve provided with you.

To recap the “see-for-yourself” proofs:

1. Proof you need to limit to one flash per refresh.
Just view http://www.testufo.com on a CRT or LightBoost display
The number of duplicate images is stroberate divided by framerate.
30fps@60Hz = double image effect
2. Proof of persistence is a motion blur problem.
– Answer 2a. Refer to Oculus Crystal Cove prototype
– Answer 2b. Refer to multiple sources confirming OLED motion blur
– Answer 2c. Refer to http://www.testufo.com/eyetracking
3. Proof of duty cycle relationship to motion blur
– See above TestUFO animations including the “Four UFO black frame insertion” animation at:
http://www.testufo.com/#test=blackframes&count=4&equalizer=1
4. Dozens of science papers are linked from:
– http://forums.blurbusters.com/viewtopic.php?f=7&t=333&start=10#p3122
– and also at https://blurbusters.com/references/

The old name “sample and hold” is a bit silly actually, since sometimes it’s curves and decay effects, rather than squarewave, but they all boil down to the same motion blur problem (persistence) caused by tracking eyes across static frames. As display transitions become faster, GtG becomes insignificant to the point of more and more intense squarewave checkerboard effect at http://www.testufo.com/eyetracking#pattern=checkerboard then persistence becomes more accurately predictable/calculatable (e.g. begins to more and more accurately follow the formula of 1ms persistence equalling 1 pixel of motion blur during 1000 pixels/second framerate=refreshrate motion). The motion blur of the checkerboard lines is equal to one frame cycle duration of persistence, blur trail exactly matches refresh rate, with GtG so insignificant that the checkerboard edges are sharp – proving the sample-and-hold effect. On slower LCDs, the checkerboard effect is more blurred when GtG is significant effect. Which shows that GtG-induced blurring has disappeared, but the persistence blur hasn’t disappeared (the checkerboard illusion itself is 100% the persistence effect itself).

Again, I point out the only two real ways to shorten persistence (frame visibility time) is either:
(A) to add increasing amount of blackness between refreshes via any method (backlight strobing, turning off OLED, phosphor decay, black frames, whatever). Doesn’t matter a hoot, doesn’t matter one iota how, as long as there’s continuous blackness between frames.
(B) to add extra intermediate frames, either via interpolation or via extra refresh cycles.

So you can see, by corollary, there’s no way to do a flickerfree method of reducing persistence without adding extra frames in any way (e.g. via interpolation, or via true refreshes). The good news is that interpolation lag becomes lower and lower the higher framerates you go, so once we hit about 240fps@240Hz, we can begin to use interpolation because it could be as little as 1/240sec or 2/240sec input lag to get to 1000fps@1000Hz or 2000fps@2000Hz or whatever high rates we might need in the future. As it stands now, interpolation (fake frames) is too laggy to be used today, but when framerates get higher, interpolation latency goes down, so once we hit a critical framerate, we can use interpolation to get the rest of the way of eliminating the rest of visible persistence without needing any form of light modulation.

in case of display reproducing variable framerate (which seems to be another great and essential but long-postponed accomplishment IMO).

We believe variable frame rate is a great thing, and especially when they raise the maximum Hz. This can provide a gradual growth towards e.g. Perfect 240fps on a flickerfree variable refresh rate display will still have 1/240sec of motion blur = 4.2ms of persistence = 4.2ms of motion blur = 4.2 pixels of motion blurring during 1000 pixels/second moition.

Well, we’ll see – if we aren’t talking about great performance gains out of silicon, but about display technology improvements, then there are still some foreseeable opportunities.

That’s why Blur Busters believe we’re stuck with various forms of light modulation (e.g. black frame insertion, phosphor decay, OLED modulation, strobe backlights) to add black periods between refreshes. So we’re huge advocates of light-modulated displays (CRT, LightBoost, low-persistence strobed OLED like Oculus, etc) and agree with the big insiders who know this stuff.

Like1Vote Down 
10 years 1 month ago
wpDiscuz

Recent Content

Blur Buster Forums

Blur Busters Forums

TestUFO Motion Tests

TestUFO Motion Tests