Balancing Pixel Response Time with Refresh Rate

[Black Octagon]

Question for you technically-competent and sexually-irresistible people out there:

To what extent does a monitor REALLY need a sufficiently-low pixel response time in order to make a higher refresh rate 'worth it'?

Longer version:
When discussing 120Hz monitors (especially overclockable IPS ones) I frequently read claims to the effect that "120Hz may be an improvement over 60Hz, but it's not as good as it could be because that panel's pixel response time is too slow to take advantage of 120Hz."

60Hz: 16.67ms per screen redraw
120Hz: 8.33ms per screen redraw
144Hz: 6.94ms per screen redraw
etc.

I interpret this as meaning that your pixel response time needs to be at least as quick as (or ideally, faster than) the time it takes for the screen to redraw. Otherwise, the pixels cannot update their colours quickly enough to for our us to see the benefit of the faster refresh rate.

1) Is this correct?
2) If it is correct, do we really need to apply this rule in terms of true/B2B pixel response time, or does it suffice for the G2G response time to be quicker than the screen redraw time?
3) Since manufacturers typically just declare a G2G and (usually) highly optimistic response time, how can we know the true B2B response time? Are reviews by the likes of PRAD.de using methodology in which we can place a lot of trust?

My current dilemma: I run an Overlord Tempest X270OC overclocked to 120Hz. The pixel response time as declared by the manufacturer is 'equal to or less than 6ms' and I fully expect this to only mean G2G. It 'feels' every bit as snappy in-game at 120fps as things did on my previous 120Hz screen - the TN-based Samsung 950D, whose (GTG) response time was 2ms - but I wonder if things would feel substantively better were the panel capable of a B2B response time less than 8ms.

Thanks in advance for the replies.

 

[Black Octagon]

I'll also accept input from anyone who is technically competent but NOT sexually irresistible, if that helps...

 

[Mark Rejhon]

To what extent does a monitor REALLY need a sufficiently-low pixel response time in order to make a higher refresh rate 'worth it'?

We need to differentiate between "pixel transition time" apart from "pixel visibility time" (true persistence factor).
There are different behaviors in strobed and non-strobed displays. To help understand the relative importance of pixel transition time, it becomes very important to understand how sample-and-hold creates eye-tracking-based motion blur. To study eye-tracking motion blur in a do-it-yourself manner, this animation comes highly recommended:

Web Animation of Sample-And-Hold Motion Blur:
www.testufo.com/#test=eyetracking

View this your LCD right now in Google Chrome or another VSYNC-supported web browser. Observe how when tracking your eyes on the moving object, the patterns show up because of motion blur. The vertical lines are smeared across your retinas because of sample-and-hold (motion blur of eye tracking). If you stop tracking your eyes, the motion blur instantly disappears and you see the vertical lines instead. Nearly all the motion blur you see in this webpage is nearly 100% caused by eye tracking, because of sample-and-hold. This is not the fault of the LCD speed, but the nature of the LCD display continuously shining for a long period (1/60sec or 1/120sec). It get blurred like a slow camera shutter (1/60sec shutter or 1/120sec shutter). Pixel transition time of modern LCD monitors is not creating the motion blur that you see in this animation above. The scientific effect is already explained in the several links above. Pixel transition speed is not important on sample-and-hold displays where the pixel transition speed is less than half a refresh. As we know, on common TN displays, we know 2ms is less than 20% of a 1/60sec refresh (1/60sec = 16.7ms). As you can already see, pixel transition time on modern LCD's is not the cause of motion blur.

Now that you understand sample-and-hold a little better, let's start to answer your question:

-- For common non-strobed displays on modern panels, long pixel visibility time is the dominant cause of motion blur, instead of pixel transition time. Pixel transition time no longer becomes important once transitions are less than about half a refresh long, because the pixel transitions are now finally hidden by sample-and-hold motion blur. For a 60Hz LCD, which continuously shines for 16.7ms, a 2ms pixel transition time is virtually completely hidden by sample-and-hold motion blur (eye-tracking-based motion blur).

-- For strobed displays (e.g. CRT, LightBoost), pixel visibility time is the chief dictator of motion blur too. On strobe-backlight LCD's, the pixel transition time (now hidden by human eye by the black period) matters for different reasons: The pixel transition time needs to fits virtually completely within the dark period (backlight off), it doesn't matter how fast the pixel transition as long as the pixel finishes transitioning before the backlight is strobed. Though faster LCD's will have less remnant effect (e.g. fainter 3D crosstalk, fainter LightBoost faint sharp ghost effect -- this is the remnant <1% incompleteness of pixel transition before the next strobe). Here, the 1ms LCD's has less 3D crosstalk (and LightBoost 2D faint-sharp-ghost effect) than the 2ms LCD's. Pixel transition time is often measured to 90% completeness, but doesn't account for the perfection of the transition. Often the pixels are gradually inching closer and closer to the final value (sometimes rippling back and fourth). This incompleteness of pixel transition is a chief cause of 3D crosstalk (faint leakage of the previous refresh into the next refresh).

Links to Pages Talking about Sample-And-Hold Motion Blur
-- Why Do Some OLED's Have Motion Blur? (same problem affects traditional LCD)
-- Michael Abrash of Valve Software explaining motion blur caused by eye tracking
(For his article, "persistence" is same thing as "sample and hold". His eye-tracking diagrams are exactly what I've often tried to explain)
-- Why We Need 1000fps @ 1000Hz This Century (uses Michael's images)
-- iD Software's John Carmack YouTube QuakeCon chat about about motion blur & strobing
For the purposes of this, "persistence", "sample-and-hold", "eye tracking motion blur" are roughly equivalent, as everyone often use different terminologies to describe similar problems.
-- Scientific Papers about Sample & Hold
-- High Speed Video of 2007 LCD; Observe how pixel transition time is slow it bleeds masively into next refresh.
-- High Speed Video of 2012 LCD; Observe how pixel transition time is fast it is mostly finished before the next refresh begins. (Makes it possible to strobe between refreshes).

 

[Firepc]

So if I understand this correctly... For a 120Hz IPS some of your response times are going to be easily beyond half of the refresh (i.e. >4ms) so they still play a role. And therefore 120Hz TN panels still are faster and will blur a bit less. But the dominant cause of motion blur is actually from eye tracking which applies to both the IPS and TN models.

 

[Mark Rejhon]

So if I understand this correctly... For a 120Hz IPS some of your response times are going to be easily beyond half of the refresh (i.e. >4ms) so they still play a role. And therefore 120Hz TN panels still are faster and will blur a bit less. But the dominant cause of motion blur is actually from eye tracking which applies to both the IPS and TN models.

Correct. The overclockable IPS displays have noticeably more motion blur than TN during 120Hz. People who owns both 120Hz overclockable IPS and 120Hz TN panels have actually reported that the 120Hz TN has noticeably less motion blur even in non-LightBoost mode, so this corroborates correctly with this too. For 5ms IPS displays, they often actually take 10ms+ for a lot of GtG values (for certain grays to certain grays). That's more than a full refresh at 1/120sec! Which means you get noticeable streaking issues. You get situations such as 3 refreshes blending into each other, similar to this:

Observe three refreshes overlapped (06, 07, and 08). This interferes with being able to get maximum achievable motion clarity. This is the TestUFO Flicker test running on an older LCD that had a 16ms response time, then photo taken at 1/1000sec exposure (single frame from high speed video). That's a fast camera exposure, yet it captures a whopping 3 different LCD refreshes! A lot of overclockable 120Hz IPS have this issue due to the slowness of the pixel transitions limiting its 120Hz benefits relative to a 120Hz LightBoost TN. Many old LCD's from the 1990's often had 100ms and 50ms response times, and even 16ms LCD's of yesterday still often had many GtG's going for more than 50ms at some times, like in this photo. That meant refreshes were overlapped, creating long streaking effects. This creates additional motion blur since the visibility of old pixels are prolonged by a very slow LCD. New refreshes were added as old refreshes faded away. So you saw heavy streaking on old LCD's. LCD's improved over the years, 33ms, 20ms, 16ms, 12ms, 10ms, 5ms, 2ms, 1ms. So, only in the last couple years, new LCD's finally became fast enough to make CRT-quality motion possible by also eliminating the sample-and-hold effect (strobe backlight. Pixel transition speed are fast enough for strobe backlight use. The backlight can finally be turned off while waiting for pixel transitions (unseen by human eyes), and the backlight is strobed only on fully-refreshed LCD frames (seen by human eyes). The strobes can be shorter than pixel transitions, breaking the pixel transition speed barrier! In addition, it eliminates the sample-and-hold effect. Motion clarity of LCD is now limitless today when we're talking about strobe-capable LCD technology. Assuming you've got a bright enough backlight for shorter strobes to compensate for the large dark period between strobe flashes. Brighter backlights needed for shorter flashes. Strobe backlights (just like CRT) behave as a sort of high speed flash photography on each completed refresh -- equivalent of completely eliminating motion blur by camera-tracking -- equivalent of eye-tracking. You also see it with strobe lights on dance floors; the dancers are no longer a blurry mess in the dark but a series of unblurred stop-motion frames. Do this fast enough at high refresh rates, and then you create the coveted "CRT effect" of perfectly sharp fast pans. And at 120Hz LightBoost, yields the CRT zero motion blur effect during perfect framerate (during locked framerate=Hz motion)

Due to the streaking, subjectively, 120Hz IPS has approximately 40% less motion blur than 60Hz due to the streaking effect of some GtG's taking longer than a refresh, while during TN, going to 120Hz manages to nearly exactly reach the theoretical (unassisted by strobe) 50% reduction in motion blur, due to the quickness of pixel transitions (not the limiting factor in motion clarity). Further improvements in motion blur is accomplished via the strobing, allowing 120Hz LightBoost (1.4ms strobes, measured by oscilloscope) to have about 12x clearer motion than a 60Hz LCD (16.7ms sample-and-hold). Where blur trails are only 1/12th as long -- e.g. 1 pixel of motion blur trailing instead of 12 pixel of motion blur trailing. The motion blur comparision graph compares these. This is faster than a Sony FW900 CRT, which has a phosphor decay longer than a LightBoost=10% strobe! (And confirmed by former FW900 CRT users who say LightBoost has less motion blur -- the first LCD with less motion blur than CRT).

Be mindful that although Lightboost TN's are supreme excellence in CRT-quality motion in an LCD panel, they do often have significantly worse colors than the 120Hz overclockable IPS. (If colors are more important than expense, avoid the VG248QE if you're willing to pay more for color. The more expensive best-color LightBoost of VG278H does reduce the compromises significantly, though, as long as you are okay with reduced brightness). The IPS/PLS panels have better colors, which can be preferred by many people, but motion blur nuts (like me) would know that there is about 6x more motion blur during regular 120Hz (sample-and-hold at 8.3ms) than with LightBoost 120Hz (strobe at 1.4ms). The overclockable IPS would have a bit more than 6x more motion blur than LightBoost, since there's still the streaking effect to consider due to the slower IPS pixel transitions. The eye tracking motion blur is only one-sixth, so the blur trail is one-sixth as long during optimized LightBoost. Which can be a very important consideration for people used to the "CRT effect" of perfectly sharp panning/scrolling with zero motion blur. You must have good stutter/tearing control because the clarity makes tearing/stutters that much easier to see. Otherwise, LightBoost can look worse because the motion blur elimination makes things so clear to the point things can look more stuttery/tearing/jittery. And some people don't track moving objects as often as other people (e.g. staring only at the crosshairs without moving eyes). So to get LightBoost motion perfection, you must eliminate all weak links in your chain, so use (1) fast GPU capable of locked framerate=Hz, 100fps@100Hz or 120fps@120Hz, *AND* (2) try using VSYNC ON or the low-lag Adaptive VSYNC (better for competitive) to get the perfect TestUFO "smooth effect" into your games; *AND* (3) good 1000Hz mouse that doesn't add micro-stutters, where you can mouse-turn left/right as perfectly smoothly as keyboard strafe left/right.

The decision to go LightBoost 120Hz or go Overclockable 120Hz, is a personal decision. Many swear by one, while others swear by the other. Different priorities (static image/color quality versus better motion quality).

 

[Firepc]

Fantastic and informative post there, a really good read. Cheers Mark!

 

[Meeho]

I always enjoy Mark's posts.

Is 120fps@120Hz more important for Lightboost LCDs than CRT, and, if so, why is it tehnically less important for CRTs?

Also, isn't Lightboost flicker worse thant CRT flicker because of phosphor retention, as PMW flicker is much faster (in Hz) than Lightboost yet way worse than >85 Hz CRT?

 

[Black Octagon]

Mark, thank you so much. I may need to re-read all this a few times to properly digest it all but already suspect you've given me the answer I was looking for and then some. interestingly, I watched Carmack's keynote earlier today and do remember his spiel about 'persistence.' It's finally starting to come together in my mind...

 

[NervousEnergy]

Great post again by Mark. I don't mind the color issues of TN, but the lack of larger formats in LB monitors is tougher. I'd love to see LB in TV sized displays with 120Hz input, but the closest you can get right now is Sony Motionflow.

 

[Mark Rejhon]

I always enjoy Mark's posts.
Is 120fps@120Hz more important for Lightboost LCDs than CRT, and, if so, why is it tehnically less important for CRTs?

LightBoost is like owning a CRT that only operates at 100-120Hz.
It wouldn't be important if LightBoost strobed at any refresh rate (it scientifically could), but nVidia/manufacturer decision to limit hardware strobing only at 100Hz-120Hz.

For best motion clarity, flicker displays such as CRT benefit the most from framerate matched with refresh rate. Flicker/strobed display eliminate so much motion blur, that stutters/tearing is much easier to see, so that literally demands framerates matched with refresh rates. That's why great Lightboost 2D demands 120fps @ 120Hz. It's merely the two combination of:
(1) LightBoost locked to strobe at high refresh rates and...
(2) Motion on strobed displays look best at framerates exactly matching refresh rates (e.g. VSYNC ON or using Adaptive VSYNC which has lower lag).
This is the combination that forces expensive GPU purchases to make LightBoost motion look the most beautiful. So yes, GPU power for 120fps@120Hz (or 100fps@100Hz) is more important than LightBoost than with CRT, because you can't lower the strobing rate like you can with CRT's. CRT's still flicker at 75Hz, but you can't do LightBoost flicker at those rates.

 

[Black Octagon]

Ok here we go. Thanks in advance for your patience. I have not undergone formal computer science training.

We need to differentiate between "pixel transition time" apart from "pixel visibility time" (true persistence factor).

Ok first thing. Is 'pixel transition time' the exact same thing as the monitor's (B2B) pixel response time or does this term refer to other factors too? More generally, what is the difference between transition and visibility time? Isn't the time it takes for a pixel to transition from one colour to the next the same thing as the amount of time for which it is 'visible'?

There are different behaviors in strobed and non-strobed displays. To help understand the relative importance of pixel transition time, it becomes very important to understand how sample-and-hold creates eye-tracking-based motion blur. To study eye-tracking motion blur in a do-it-yourself manner, this animation comes highly recommended:

Web Animation of Sample-And-Hold Motion Blur:
www.testufo.com/#test=eyetracking[/size]

View this your LCD right now in Google Chrome or another VSYNC-supported web browser. Observe how when tracking your eyes on the moving object, the patterns show up because of motion blur. The vertical lines are smeared across your retinas because of sample-and-hold (motion blur of eye tracking). If you stop tracking your eyes, the motion blur instantly disappears and you see the vertical lines instead. Nearly all the motion blur you see in this webpage is nearly 100% caused by eye tracking, because of sample-and-hold. This is not the fault of the LCD speed, but the nature of the LCD display continuously shining for a long period (1/60sec or 1/120sec). It get blurred like a slow camera shutter (1/60sec shutter or 1/120sec shutter). Pixel transition time of modern LCD monitors is not creating the motion blur that you see in this animation above.

Ok so far so good, I think. I very much like the UFO Test link. I also notice that the extent of the motion blur seems to vary slightly depending on the illusion I select.

Pixel transition speed is not important on sample-and-hold displays where the pixel transition speed is less than half a refresh. As we know, on common TN displays, we know 2ms is less than 20% of a 1/60sec refresh (1/60sec = 16.7ms). As you can already see, pixel transition time on modern LCD's is not the cause of motion blur.

(snip)

For common non-strobed displays on modern panels, long pixel visibility time is the dominant cause of motion blur, instead of pixel transition time. Pixel transition time no longer becomes important once transitions are less than about half a refresh long, because the pixel transitions are now finally hidden by sample-and-hold motion blur. For a 60Hz LCD, which continuously shines for 16.7ms, a 2ms pixel transition time is virtually completely hidden by sample-and-hold motion blur (eye-tracking-based motion blur).

Ok, so if the pixel transition speed of a 60Hz display is less than approximately 8ms then the display itself shouldn't create substantive motion blur. However, doesn't this imply that higher refresh rate displays need even FASTER pixel transition times in order to stave off motion blur?? I.e., a refresh takes 8.33ms on a 120Hz display, meaning that you would need a pixel transition speed less than approximately 4ms?! I suspect I'm still confused on this point...

Pixel transition time is often measured to 90% completeness, but doesn't account for the perfection of the transition. Often the pixels are gradually inching closer and closer to the final value (sometimes rippling back and fourth). This incompleteness of pixel transition is a chief cause of 3D crosstalk (faint leakage of the previous refresh into the next refresh).

So - again, assuming that pixel transition time = B2B pixel response time - are we talking here about black-to-black or grey-to-grey? And when you refer to this being 'measured' can we take much confidence in the measurements of PRAD.de and co.? I ask this because we're often warned to read display manufacturers' declared response times (in monitor tech specs) with a grain of salt.

Links to Pages Talking about Sample-And-Hold Motion Blur
-- Why Do Some OLED's Have Motion Blur? (same problem affects traditional LCD)
-- Michael Abrash of Valve Software explaining motion blur caused by eye tracking
(For his article, "persistence" is same thing as "sample and hold". His eye-tracking diagrams are exactly what I've often tried to explain)
-- Why We Need 1000fps @ 1000Hz This Century (uses Michael's images)
-- iD Software's John Carmack YouTube QuakeCon chat about about motion blur & strobing
For the purposes of this, "persistence", "sample-and-hold", "eye tracking motion blur" are roughly equivalent, as everyone often use different terminologies to describe similar problems.
-- Scientific Papers about Sample & Hold
-- High Speed Video of 2007 LCD; Observe how pixel transition time is slow it bleeds masively into next refresh.
-- High Speed Video of 2012 LCD; Observe how pixel transition time is fast it is mostly finished before the next refresh begins. (Makes it possible to strobe between refreshes).

Thanks a lot. The final 30 seconds of that Asus VG278H video is seriously dangerous for my wallet

In the Sluyterman paper, when he refers to 'hold time' is this the same thing as pixel visibility time?

 

[Black Octagon]

Correct. The overclockable IPS displays have noticeably more motion blur than TN during 120Hz. People who owns both 120Hz overclockable IPS and 120Hz TN panels have actually reported that the 120Hz TN has noticeably less motion blur even in non-LightBoost mode, so this corroborates correctly with this too.

Ok, I do need to chime in here. I went directly from a non-LB 120Hz native TN display (Samsung S27A950D) to a 120Hz overclocked IPS (Tempest X270OC) and can honestly say that I find the motion blur and overall 'snappiness' of the displays at 120Hz to be identical, even when run side-by-side...

Maybe I'm not so sensitive to motion blur or perhaps my eyes aren't great, but I do notice a massive difference between 60Hz and 120Hz so they can't be so far gone.

Perhaps the Samsung, despite being 120Hz, simply has higher latency than some of the alternative 120Hz TNs from BenQ or ASUS? I certainly read this before purchasing my Samsung, and only chose it over a BenQ or ASUS because it was regarded as having particularly good colour for a TN.

For 5ms IPS displays, they often actually take 10ms+ for a lot of GtG values (for certain grays to certain grays). That's more than a full refresh at 1/120sec! Which means you get noticeable streaking issues. You get situations such as 3 refreshes blending into each other, similar to this:

Observe three refreshes overlapped (06, 07, and 08). This interferes with being able to get maximum achievable motion clarity. This is the TestUFO Flicker test running on an older LCD that had a 16ms response time, then photo taken at 1/1000sec exposure (single frame from high speed video). That's a fast camera exposure, yet it captures a whopping 3 different LCD refreshes! A lot of overclockable 120Hz IPS have this issue due to the slowness of the pixel transitions limiting its 120Hz benefits relative to a 120Hz LightBoost TN.

Hmm, as I said in my OP, the manufacturer of my overclocked 120Hz IPS declares the pixel response time as being 'equal to or less than 6ms.' Are you saying that this is only true for certain GTG transitions and for others (and definitely B2B transitions) the time is a lot longer? Has this actually been measured? To my knowledge these panels haven't undergone any professional reviews, at least not when run at 120Hz.

LCD's improved over the years, 33ms, 20ms, 16ms, 12ms, 10ms, 5ms, 2ms, 1ms. So, only in the last couple years, new LCD's finally became fast enough to make CRT-quality motion possible by also eliminating the sample-and-hold effect (strobe backlight. Pixel transition speed are fast enough for strobe backlight use. The backlight can finally be turned off while waiting for pixel transitions (unseen by human eyes), and the backlight is strobed only on fully-refreshed LCD frames (seen by human eyes). The strobes can be shorter than pixel transitions, breaking the pixel transition speed barrier! In addition, it eliminates the sample-and-hold effect. Motion clarity of LCD is now limitless today when we're talking about strobe-capable LCD technology. Assuming you've got a bright enough backlight for shorter strobes to compensate for the large dark period between strobe flashes.

Mark, as someone who's been following this so intensively, do you anticipate that strobing backlights will soon become the 'norm' for LCDs? Any reason why this would or would not be possible on displays whose (TN) panels have a higher than 1920x1080 resolution? I went for a 120Hz overclocked display partially because I wasn't aware of LightBoost at the time, but even if I was I'm still not entirely sure I would have chosen different. The color of IPS is one obvious factor, but so was the 2560x1440 resolution, which I must say I find to be hugely attractive on a 27-inch display.

The decision to go LightBoost 120Hz or go Overclockable 120Hz, is a personal decision. Many swear by one, while others swear by the other. Different priorities (static image/color quality versus better motion quality).

I do not regret my decision to go the overclockable 120Hz IPS route, but do get curious about installing a LB 120Hz monitor alongside it purely for FPS gaming.

Thanks again Mark!

 

[Black Octagon]

bump

 

[Mark Rejhon]

Ok first thing. Is 'pixel transition time' the exact same thing as the monitor's (B2B) pixel response time or does this term refer to other factors too? More generally, what is the difference between transition and visibility time?

Pixel transition time is the time it takes for the pixel to go from one color value to the next. Also known as GtG, or BWB, or WBW, or B2W, or W2B, all of them involve a transition from a pixel state to a different pixel state, or involves a transition from a pixel state to another one and then back. The pixel doesn't disappear, the pixel persists afterwards to continue to interfere with motion blur via the sample-and-hold effect (which is a motion blur factor independent of pixel transition time).

Persistence (aka "hold time", "pixel visibility time", aka "sample-and-hold") is the amount of time the pixel is "held" at the same color for, before it essentially disappears or becomes replaced by the next frame. This is the true measurement of motion blur. Shorter persistence will lead to less motion blur. The longer a visible frame remains static, the more opportunity for the frame to gets blurred across your vision during eye-tracking, as has already been explained. Measurements such as MPRT (Motion Picture Response Time) takes into account of persistence.

Ok so far so good, I think. I very much like the UFO Test link. I also notice that the extent of the motion blur seems to vary slightly depending on the illusion I select.

They can diverge if BWB and WBW transition times are different. The first pattern (stars) creates the stars optical illusions out of WBW transitions, while the final pattern (lines) creates the optical illusion out of BWB transitions, at www.testufo.com/#test=eyetracking

Ok, so if the pixel transition speed of a 60Hz display is less than approximately 8ms then the display itself shouldn't create substantive motion blur. However, doesn't this imply that higher refresh rate displays need even FASTER pixel transition times in order to stave off motion blur?? I.e., a refresh takes 8.33ms on a 120Hz display, meaning that you would need a pixel transition speed less than approximately 4ms?!

That's correct. It's not a hard-and-fast number but generally, that's the crossover point where more motion blur begins to be created by the sample-and-hold effect, than motion blur created by the slow pixel transition. So on fast TN displays, the dominant cause of motion blur is the sample-and-hold effect. This is also why LightBoost makes such a dramatic improvement in motion clarity, since the LCD's potential is being bottlenecked by the sample-and-hold effect.

So - again, assuming that pixel transition time = B2B pixel response time - are we talking here about black-to-black or grey-to-grey? And when you refer to this being 'measured' can we take much confidence in the measurements of PRAD.de and co.? I ask this because we're often warned to read display manufacturers' declared response times (in monitor tech specs) with a grain of salt.

Pixel transition times are very useful measurements. But they become less and less useful as we go smaller than one frame cycle. Old software such as PixPerAn (circa 2001, over ten years old) are not aware of the use of strobing on LCD's to reduce motion blur, so new tools are now needed.

Pixel response time less than half a frame long, means monitors are so identical in motion blur that it is hard to tell them apart in motion clarity (very subtle differences). The sample-and-hold is the bottleneck today. Further motion clarity improvement needs to be done by shortening the hold time (either by more Hz, or by adding black gaps between Hz). Obviously, it's easier to strobe than to go to a true-240Hz or true-480Hz LCD. (and we still dream of the 1000fps@1000Hz display to try to bypass most of the diminishing returns "most of the way" to Holodeck league).

In the Sluyterman paper, when he refers to 'hold time' is this the same thing as pixel visibility time?

Correct. "Hold time" is the same thing as "pixel visibility time".

 

[Mark Rejhon]

Ok, I do need to chime in here. I went directly from a non-LB 120Hz native TN display (Samsung S27A950D) to a 120Hz overclocked IPS (Tempest X270OC) and can honestly say that I find the motion blur and overall 'snappiness' of the displays at 120Hz to be identical, even when run side-by-side...

The difference is only about 10% -- it is extremely subtle and can only easily be noticed during perfect framerate-locked motion (e.g. VSYNC ON) of high-contrast edges, or in tests such as PixPerAn or TestUFO. The difference is easily lost in microstutters and/or during VSYNC OFF motion. If you still yave both displays, view www.testufo.com/#test=ghosting

as I said in my OP, the manufacturer of my overclocked 120Hz IPS declares the pixel response time as being 'equal to or less than 6ms.' Are you saying that this is only true for certain GTG transitions and for others (and definitely B2B transitions) the time is a lot longer?

Correct, some values can be a lot longer. It may be truly 6ms or less to a 90% completeness, but what about the pixel transition time to a 99% completeness? (Even 99% complete still creates a bit of 3D crosstalk or LightBoost double-image effect). Pixel transitions don't settle at the final value perfectly stable, it can get closer and closer gradually in a logarithmic curve, or during overdrive, it can ripple back and fourth, creating overdrive artifacts. (e.g. when transitioning to a 50% gray, it may overshoot back and fourth between a 45% and 55% gray over a few milliseconds before stabilizing on a 50% gray).

Pixel transition time is not as simple as an exact value. Human visible artifacts still remain long after the manufacturers rated time, due to various factors. Faint artifacts can still be visible 20ms later even on a 2ms LCD, under certain worst-case-scenario circumstances.

Has this actually been measured? To my knowledge these panels haven't undergone any professional reviews, at least not when run at 120Hz.

There's dozens of graphs throughout if you dig deep. But I have a job & I don't have time to search down all the hyperlinks and detail right now [apologies] -- in one of the Samsung factory papers, there's a 3D bar graph where X axis is the original pixel value and the Y axis is the destination pixel value, and the Z axis is the number of milliseconds. It's a very hilly 3D bargraph. Some scientific papers include such a 3D bargraph. Plus even that, if you isolate to one bar, and display the pixel value for one GtG gransition along the axis of time (as the milliseconds tick by), you will see the pixel is never perfectly stable, as if it's teeter-tottering around its final pixel value. How do you nail an exact pixel value then; you define a cutoff point where it's considered practically complete. (e.g. "90% completeness cutoff").

Mark, as someone who's been following this so intensively, do you anticipate that strobing backlights will soon become the 'norm' for LCDs? Any reason why this would or would not be possible on displays whose (TN) panels have a higher than 1920x1080 resolution? I went for a 120Hz overclocked display partially because I wasn't aware of LightBoost at the time, but even if I was I'm still not entirely sure I would have chosen different. The color of IPS is one obvious factor, but so was the 2560x1440 resolution, which I must say I find to be hugely attractive on a 27-inch display.

It's hard to say, but could take several years. We need to encourage more manufacturers to implement this technology now. Past attempts (e.g. BENQ AMA-Z in 2006) did not succeed as well, but today, with high-efficiency strobe backlights such as LightBoost are now finally possible on LCD's, the time is good right now to implement such technology where possible.


EDIT: I spent the time and found a few graphs for you:

3D bar graph for GtG times.

3D bar graph for overdrive overshoots (pixel bounce).

Good strobe backlights (with zero faint-ghost effect or crosstalk effect) demand perfectly-settled transitions with nearly zero error, and that's a tall order for LCD panels. It may be truly 6ms to "good enough for a sample-and-hold display", but it's really often "10ms or 20ms" to the levels necessary to be clean enough for a strobe backlight. Even the 1ms/2ms panels don't always produce perfectly clean strobes, since often a little remnant (1%) is still visible over 8ms later; creating the 3D crosstalk (or 2D faint-sharp-ghost effect). The crosstalk between refreshes is the faint remnant of the previous refresh leaking into the next refresh. It's also seen in the high speed video at http://www.youtube.com/watch?v=hD5gjAs1A2s where beginning at 0:55 afterwards, you will see extremely ultra-faint afterimages from the previous refresh leaking into the next refresh (~1% intensity). This is the factor that creates 3D crosstalk (and when used in 2D, the LightBoost 2D sharp trailing faint ghost double-image effect)

Right now, I think I've filled the single-person-answering quota for the day (LOL). Blur Busters probably will have to create an article related to this at some point, to help explain to people about pixel response time behaviours, at some point... (I believe Adam of pcmonitors.info wanted to do this, so I'll check on him too)

 

[Black Octagon]

Wow, thankyou so much. You've really helped me understand all this better. FYI: if you published a text book on monitor technology for dummies I would buy it without hesitation

 

[Black Octagon]

Having now continued a little with my own newb research, I've finally stumbled across a (highly basic) tidbit on which I was previously confused. I've read that the terms 'motion blur' and 'ghosting' are actually interchangeable when talking about display performance (not when talking about motion blur as an intentionally-created cinema effect, obviously).

These interchangeable terms both refer to an on-screen 'streaking' effect that results from when one tries to display fast motion on a display with a 'slow' response time. Yes, this is basic stuff to some of you but TBH I was getting stumped by my assumption that each bit of terminology referred to a different effect.

So anyway, based on this I am concluding that having a high refresh rate on a display with insufficient pixel transition time (particularly a transition time that is insufficient because of an excessive amount of 'persistence' / 'hold time' / 'pixel visibility time' / 'sample-and-hold') causes motion blur / ghosting.

In other words, overclocking an IPS display from 60Hz to 120Hz does increase fluidity of motion but also leads to streaking of a rapidly-moving picture as a side-effect.

I must say, I don't seem to notice this on my own Tempest @ 120Hz but this is perhaps due to the fact that I've been on LCDs for years and haven't gamed on a quality CRT since my university days. Perhaps if I had gone straight from my old Sony Trinitron to my overclocked IPS I'd be less happy with it than I currently am.

Now, the only remaining challenge is to keep my interest in all this stuff purely 'academic.' Part of me itches to try out LightBoost but another part of me does not want to shell out the money needed right now. And yet another part of me does not want to experience something that will destroy my ongoing honeymoon with my Tempest, lol

 

[geok1ng]

Here, the 1ms LCD's has less 3D crosstalk (and LightBoost 2D faint-sharp-ghost effect) than the 2ms LCD's.

I do not mean to troll you mark, but this kind of statement is an irresistible troll bait.

even if one day we have TRUE b2B 1ms pixel response times, i do think that someone would launch a 1million dollars prize to anyone that could tell the difference from 2ms response times.

Heck, even fake 2ms vs fake 5ms is pretty much impossible to tell apart in apples to apples situations. We are talking 2 orders of magnitude below human reaction time and an order of magnitude below human perception time. I do understand you cruzade for strobed monitors, but this 1ms better than 2ms stuff is weird, to say the least.

Why we need 1000fps at 1000hz is a tech goal similar to "why we need to clone monica sweetheart so every woman looks like her" but with much less real life reasons to justify.

 

[Meeho]

We are talking 2 orders of magnitude below human reaction time and an order of magnitude below human perception time.

Oh?! I don't think so.

 

[Mark Rejhon]

even if one day we have TRUE b2B 1ms pixel response times, i do think that someone would launch a 1million dollars prize to anyone that could tell the difference from 2ms response times.

Before I begin to reply:
1. Have you heard of something called "3D crosstalk"?
2. Do you know what 3D crosstalk is caused by? (Hint: For strobed LCD's, the dominant cause isn't the shutter glasses)

3D crosstalk is the faint remnant pixel leakage between frames.
I generally agree with you when we're talking where the pixel transition speed limitations are fully hidden by sample-and-hold motion blur.

...But the ballgame is different when it comes to 3D / strobing -- which is used for 3D and 2D -- So I have nothing further to say to you geo1king, if you have no scientific understanding of what 3D crosstalk is, and have never seen how much better the 3D crosstalk is on the 1ms panels over the 2ms panels, unless you've seen it with your own eyes. With strobing the motion blur of sample-and-hold is eliminated, so the noisefloor for detecting remnant pixel transitions (AKA 3D CROSSTALK) is greatly lowered.

Yes, the 'fakeness' of manufacturer specs is a legitimate call out, though generally the 1ms panels usually still tend to be faster on average than the 2ms panels, even in real-world measured numbers (e.g. instead of 1ms vs 2ms, it might become more like 3ms vs 5ms for a specific GtG to a 99% transition completeness factor); and the quicker-to-complete translates to less 3D crosstalk. Even a pixel color off by less than 0.5% is detected easily as 3D crosstalk between bright edge boundaries, like between a super bright sharp color edge against a very dark color. A 2ms panel may successfully say, transition to about 1% of the final color value (more intense 3D crosstalk) while the 1ms panel may successfully, say, transition to less than 0.5% of the final color value (less 3D crosstalk).

I hereby tell you to go ahead and put up a bet against me: The science of 3D crosstalk is well sound. The occasionally detected (as occasional as 3D crosstalk) 2D LightBoost faint double image effect (described as "ghosting", but not of the traditional blurred ghosting variety) -- a faint non-blurred perfectly-sharp double image that trails behind the moving object in the same motion vector -- and detected by the same color boundaries that 3D crosstalk is detected in -- and is caused by exactly identical reasons to 3D crosstalk & identical in faintness to 3D crosstalk on exactly the same color values, since it's no longer masked by the sample-and-hold motion blur.

Not saying the 1ms panels are necessarily better in other attributes, as color quality of other (slower) panels can be vastly superior (e.g. IPS). The 27" models (e.g. VG278H, XL2720T) is now known by several LightBoost users to have better colors than the 1ms models (VG248QE, XL2411T, XL2420TE) but slightly worse crosstalk. Vega has said he clearly sees the increased crosstalk but that the better LightBoost colors outweighs this, and switched his rig from a triple-VG248QE surround into a triple-XL2720T surround.

But that is besides the point; the point is that the human detectability of the faster panels is actually detectable as less stereoscopic 3D crosstalk (and 2D trailing sharp-ghost is caused by the same reasons).