Blur Busters Forums

I recently learned about the Leo Bodnar input lag tester-- but the values I've seen for Leo Bodnar seem really out of whack with photo testing. Are Bodnar numbers accurate? Here are the things that make me suspicious:

1) Everybody always said that CRTs have no display lag (except from the scan line). This was because CRTs didn't have miles of circuitry for electrons to travel or any memory for buffers, I believe-- so the only possible lag could have been from the speed of electrons, which is high. But then Leo Bodnar tests came out and almost universally saw larger input lags than CRT comparisons did, and it looks like the explanations changed: the reason Bodnar numbers are higher is because they include lag that would be ignored with a comparison.

2) Although most Bodnar numbers are higher than the historical CRT comparison, a few are suspiciously low. Look at http://images.anandtech.com/graphs/graph7931/62611.png which is a chart comparing input lag, as measured by Bodnar tester, across monitors. Supposedly, it is an average value-- which I take to mean that it averages top, middle, and bottom of the screen. But the MX299Q is a 60hz monitor. Even with no extra lag, scan line lag should mean an average of 8.33 ms input lag. More, if previous comparison were flawed because they ignored the lag of comparison displays, then monitors with low lags by Bodnar should have been showing negative input lag, and they never did.

I would be less suspicious if Bodnar's tester wasn't such a clear convenience, or if those using it went into more detail with their methodology. Is it just a situation where poorly collected numbers are being compared to a new set of poorly collected numbers?

I confirm that Bodnar is usually fairly accurate.
With a huge "BUT"

However, it is a matter of perspective, because it is somewhat of a blackbox that hides some variables:

For comparision's sake, Bodnar TOP value is more directly comparable to photo testing.
There are SO many error factors in input lag measurements.

-- Are you measuring relative input lag measurements? Relative to what? Relative to top of scanout? Relative to video output? Relative to a different display etc? Whole chain measurements (mouse-button-to-photons)? etc.
-- Photo testing measures scan differential between displays
-- False alarms/late response/early response from transients and different lag for different colors/different brightnesses
.......PWM dimming can add an error factor. 180Hz unsynchronized PWM, at low brightness settings, adds a 1/180sec error to input lag measurements, due to the black gaps between image visibility.
.......Plasma subfields, DLP dithering, can occur earlier and at a higher frequency, during brighter colors than during dim colors (e.g. bright colors can have lower lag than dim colors on SOME (but not all) of those types of displays).
.......SMTT uses yellow/blue transitions (displaying yellow numbers on a blue screen) while Leo uses black/white flashing squares. The different input lag of different colors are a known scientific error factor. GtG is different speeds for different color pairs; impacting lag.
-- Scan can behave differently for plasma displays/strobe backlights (global presentation), versus CRT and LCD (top-to-bottom scan presentation).
.......Plasma and LightBoost has same input lag (relative to VSYNC) for top/bottom
.......CRT and LCD have different input lag (relative to VSYNC) for top/bottom
-- Factoring in GtG hugely complicates input lag calculations; it wreaks havoc in properly estimating input lag.
.......Do you measure to first pixel visibility? (e.g. 10% visible? Which is obviously human visible)
.......Do you measure to 50% pixel visibility? (e.g. halfway through GtG?)
.......Do you measure to 100% pixel visibility? (e.g. GtG fully finished)
.......It is unknown at which point Leo Bodnar terminates the lag measurement, but I would suggest midpoint of GtG as the ideal value.
.......Estimating GtG from a photo is different from estimating GtG from electronic lag testing; SMTT numbers are very visible in photo tests when GtG is less than 50% complete, so when GtG (e.g. 8ms) is a significant error factor, latency measurement error can be quite huge.
.......For simplicity, some sites such as TFTCentral just adds half the GtG measurement to input lag totals.
- When electronically measuring lag (Leo Bodnar style), do you begin measuring from beginning of VSYNC? Start/end of Front Porch? Start/end of Back Porch? First active scan line? I don't know at which scanline the Leo begins its input lag measurement clock at; the VSYNC interval can be about 0.5ms-1.0ms -- e.g. 1125 lines at 1080p means 1080/1125ths of a 60Hz refresh cycle is active, while 45/1125ths of 1/60sec is an idle period between refreshes (that's about 2/3rd of a millisecond error by not knowing this datapoint).

2) Although most Bodnar numbers are higher than the historical CRT comparison, a few are suspiciously low. Look at http://images.anandtech.com/graphs/graph7931/62611.png which is a chart comparing input lag, as measured by Bodnar tester, across monitors. Supposedly, it is an average value-- which I take to mean that it averages top, middle, and bottom of the screen. But the MX299Q is a 60hz monitor.

I am not sure, but as you can see above, there are a lot of potential mathematic error factors.
This can be caused by as something as simple as GtG differences. A low-lag but slow GtG, might mean you've got very low input lag (e.g. 20ms to first pixel visibility, but 35ms to full GtG completion). e.g. a 16ms slow-transition display that is fast during the first 50% of a GtG, may have lower lag than a 8ms faster-transition display that is really slow during the first 50% of a GtG. And as all us LCD afficanados know, GtG can be very different for different color pairs; and the different input lag measurement methods measure lag from different colors (e.g. SMTT blue/yellow transition versus Leo black/white transition)

Industry standardization suggests that lag should be measured to 50% pixel transition midpoint. However, humans can detect a pixel being visible less than 50% of its transition -- e.g. a 25% transition from black to white, is still a visible dark gray, which could still occur in the first 1-2ms of an 8ms pixel transition (IPS panel). Now, there's nonlinearities in GtG pixel transitions (e.g. logarithmic curves), that wreak havoc on the error margin of an input lag measurement.

I would be less suspicious if Bodnar's tester wasn't such a clear convenience, or if those using it went into more detail with their methodology. Is it just a situation where poorly collected numbers are being compared to a new set of poorly collected numbers?

Na, it's just Leo Bodnar simply being a blackbox with a number of unknown factors (e.g. which scanline it begins the timer at) (e.g. does it measure to GtG midpoint?). It's accurate, just that it's mathematically hard to compare the input lag numbers of all the different input lag measurement methods.

Thanks for the in-depth answer, I figured this was the place I could get details

There was something stupid I did-- I was looking at some wrong numbers, and I thought the Asus monitor I mentioned was listed with a 4ms lag. The chart actually lists 9 something ms, so it's not crazily low, just good.

What about using HDMI->DVI adapters? Can DVI derived values be compared accurately with HDMI derived values?

And....room temperature! LCD's have much slower GtG when cold than when warmed up. Sometimes GtG is by 2-3x slower when extremely cold -- e.g. left in freezing temperatures then brought in. (especially VA LCDs).

The ASUS VG248QE I have sitting here, has about 3-4ms lag in my manual photodiode oscilloscope test, from the computer-side timing of Direct3D Present() to first visible pixels showing up at the top edge of the screen. Most recent TN 120Hz monitors (in Instant mode, non-strobe mode) have roughly this ballpark of input lag, when factoring in the video cable transmission and GtG. (GSYNC modified, but photodiode oscilloscope tested running in non-GSYNC mode)

Yes, the different connections; HDMI, DVI, DisplayPort, and VGA, can create differences of a millisecond or two; VGA is the lowest lag, while HDMI/DVI/DisplayPort has a bit more. You will have to add a ~2ms rounding error when comparing against different cable technologies. Add more error margin; if going through electronic switches -- added lag (+1ms or a smidgen more) can occur when going through repeaters, switches, and surround sound receivers.

Yeah, funny thing: Some old VA-monitors with energy intensive CCFL backlight show less motion blur after some time than newer ones with "colder" LED backlight.

aufkrawall wrote:Yeah, funny thing: Some old VA-monitors with energy intensive CCFL backlight show less motion blur after some time than newer ones with "colder" LED backlight.

Yes, the CCFL heats up the VA LCD, to make the panel respond faster.
So it takes much longer for a LED-backlit VA LCD to warm up.

That said, Turbo 240 does considerably reduce VA motion blur by a huge amount, so that a Turbo240-enhanced VA LCD (even LED backlit), can have less overall motion blur than a TN/IPS LCD that's not using a strobed backlight. But you need to let it warm up a bit, though!

I had that particular Eizo monitor.
Picture was sharp in motion, but unfortunately, there were excessive color mix ups. VA panels always have some very high single response times.
Thus I'd like to see an IPS monitor with 120Hz + strobing. But no one seems to care about it...

aufkrawall wrote:I had that particular Eizo monitor.
Picture was sharp in motion, but unfortunately, there were excessive color mix ups. VA panels always have some very high single response times.

VA panels certianly have much slower response. That said, it isn't that bad -- the Eizo monitor actually gets much better when warmed up for 30-45 minutes.

Cold Eizo FG2421:



Warm Eizo FG2421:



So make sure your FG2421 warms up first, to get the best motion quality.

afair, it also happened once it was warm.
Try moving around this image in native resolution:
http://cdn.overclock.net/8/8a/8a4a5d21_ ... _zozvk.png

Chief Blur Buster wrote:VGA is the lowest lag, while HDMI/DVI/DisplayPort has a bit more.

Are there actual measurements of this? or is it something related to the HDMI/DVI/DisplayPort protocols?
I suppose it varies depending on the graphics card and monitor...