Display Testing Explained: How We Test Monitors and TVs

Viewing Angles, Uniformity, Pixel Response and Input Lag

Viewing Angles

The more monitors we test, the more we can see that off-axis viewing performance is dependent not only on pixel structure (IPS, PLS, TN, and so on), but backlight technology as well. The anti-glare layer makes a difference too.

In this test, a picture is worth one thousand words. We set a Panasonic Lumix camera to manual exposure and zero its white balance to each individual monitor. No settings are changed between shots. The top and side photos are taken at a 45-degree angle off-axis. Then, the three images are assembled into a composite. It’s a good approximation of what the eye actually sees when viewing a monitor off-center.

  • Patterns used: Gray Steps (horizontal and vertical)
  • Panasonic Lumix DMC-LX7, manual exposure
  • Off-axis angle: 45 degrees horizontal and vertical

Screen Uniformity: Luminance

To measure screen uniformity, zero and 100-percent full-field patterns are used, and nine points are sampled. First, we establish a baseline measurement at the center of each screen. The surrounding eight points are then measured. Their values get expressed as a percentage of the baseline, either above or below. This number is averaged. It's important to remember that we only test the review sample each vendor sends us. Other examples of the same monitor can measure differently.

Black field uniformity is also known as light bleed. If it’s visible, it shows up as light areas on an otherwise black screen. If the value is under 10 percent, we consider the monitor essentially perfect with no visible problems.

In the white field test, our benchmark is the same 10 percent. Few displays score higher than that.

If a display has a uniformity compensation feature, we run the test with it off and on and compare the results.

  • Patterns used: 100-percent White Field, 0-percent Black Field
  • Where appropriate, we compare measurements with Uniformity Compensation On and Off
  • Results under 10-percent mean no aberrations are visible to the naked eye

Screen Uniformity: Color

To measure color uniformity, we display an 80-percent white field and measure the Delta E error of the same nine points on the screen. Then we subtract the lowest value from the highest to arrive at the result. A smaller number means a display is more uniform. Any value below three means a variation that is invisible to the naked eye.

As in the white uniformity test, it’s rare that a display shows any color shift from zone to zone. The larger screens of HDTVs are more susceptible to errors here.

  • Pattern used: 80-percent Gray Field
  • Where appropriate, we compare measurements with Uniformity Compensation On and Off
  • Results under 3 Delta E mean no color shift is visible to the naked eye

The Tests: Pixel Response and Input Lag

To perform these tests, we use a high-speed camera that shoots at 1000 frames per second. Analyzing the video frame-by-frame allows us to observe the exact time it takes to go from a zero-percent signal to a 100-percent white field.

The pattern generator is placed at the base of the monitor so our camera can capture the precise moment its front-panel LED lights up, indicating that a video signal is being received. With this camera placement, we can easily see how long it takes to fully display a pattern after pressing the button on the generator’s remote. This testing methodology allows for accurate and repeatable results when comparing panels.

Here’s a shot of our test setup. Click on the photo to enlarge.

The brighter section of the camera’s screen is what appears in the video. You can see the lights of the pattern generator in the bottom of the viewfinder. We flash the pattern on and off five times and average the results.

When we test monitors with refresh rates greater than 60 Hz, we have to use a PC as the signal source. We use the same white field pattern and trigger its appearance with a mouse movement. That motion is recorded with the high-speed camera to see precisely how long it takes the screen to fully render after the mouse is moved. This test method is also run five times and the result averaged.

The response chart shows only how long it takes for the panel to draw a full white field from a black screen. To calculate the total input lag, we first time the period between initiating the signal to the beginning of the refresh cycle. Then we add the screen draw time to arrive at the final result.

  • Pattern used: 100-percent White Field
  • Camera: Casio Exilim EX-ZR100 set to 1000 fps (1 frame = 1 millisecond)
  • Video analyzed frame-by-frame to calculate result
Create a new thread in the US Reviews comments forum about this subject
This thread is closed for comments
11 comments
    Your comment
  • Chetou
    Thank you for this writeup. Though I do find 200 cd/m2 retina scorching on any monitor and viewing conditions, especially with large screens.

    Do you take your measurements in a dimmed/dark room? Even with the meter flush with screen, some light can pass through the glass on the sides.
    2
  • Chetou
    double post
    -2
  • cpm1984
    Gotta ask about your 'total lag' measurements - they seem to be generally much higher than what TFT Central measures. For example, the Dell UP3214Q has 97ms total lag with Toms':
    http://www.tomshardware.com/reviews/benq-pg2401pt-24-inch-monitor,3848-10.html

    but it has only 29ms lag at TFT Central (and only 25ms in 'game mode'):
    http://www.tftcentral.co.uk/reviews/dell_up3214q.htm

    Most of the monitors you review seems to have total lag around 80-100ms, which seems really slow. Slow enough that you'd notice the mouse lagging when you move it around. Yet I can't feel any appreciable lag on my Dell 2713HM (but I have no way of measuring....)
    4
  • cpm1984
    Gotta ask about your 'total lag' measurements - they seem to be generally much higher than what TFT Central measures. For example, the Dell UP3214Q has 97ms total lag with Toms':
    http://www.tomshardware.com/reviews/benq-pg2401pt-24-inch-monitor,3848-10.html

    but it has only 29ms lag at TFT Central (and only 25ms in 'game mode'):
    http://www.tftcentral.co.uk/reviews/dell_up3214q.htm

    Most of the monitors you review seems to have total lag around 80-100ms, which seems really slow. Slow enough that you'd notice the mouse lagging when you move it around. Yet I can't feel any appreciable lag on my Dell 2713HM (but I have no way of measuring....)
    -1
  • ceberle
    Anonymous said:
    Gotta ask about your 'total lag' measurements - they seem to be generally much higher than what TFT Central measures. For example, the Dell UP3214Q has 97ms total lag with Toms':
    http://www.tomshardware.com/reviews/benq-pg2401pt-24-inch-monitor,3848-10.html

    but it has only 29ms lag at TFT Central (and only 25ms in 'game mode'):
    http://www.tftcentral.co.uk/reviews/dell_up3214q.htm

    Most of the monitors you review seems to have total lag around 80-100ms, which seems really slow. Slow enough that you'd notice the mouse lagging when you move it around. Yet I can't feel any appreciable lag on my Dell 2713HM (but I have no way of measuring....)


    TFT Central uses the SMTT software which measures only the actual display lag. Our test indicates the total time it takes for a user input to translate to the screen. We account for the lag inherent in the input source which in our case is a pattern generator. While this device may be slower than the average mouse or keyboard it is completely consistent and repeatable. If we used a mouse-to-PC test, timing could shift if a driver were updated or we changed video boards. Every monitor we've measured over the past two years has been tested with the same Accupel signal generator.

    Our principal goal is consistency from monitor to monitor. By having the exact same equipment and test parameters, we can ensure that results from last year are comparable to more recent tests.

    -Christian-
    1
  • KevinAr18
    The response time testing method is done wrong and is very misleading. The tesst do not always reflect the real response of the monitors.

    Why?
    Monitors have thousands of different response times, not a single number. Testing just one or two transitions can easily give very innaccurate results; the one transition you tested may be much slower or faster than the other thousand or more transitions that exist. In order to find out if the monitor is any good, you must test a wide range of transitions.

    Want proof?
    See:
    http://www.xbitlabs.com/articles/monitors/display/viewsonic-fuhzion-vx2268wm_4.html#sect1
    In "standard" mode, the black to white transition is low, while almost all the others are high. (Note: the 255 to 0 transiyon is hidden behind the higher ones.) If you had tested this monitor using your current method, you would have concluded it is VERY fast, when, in fact, "standard" mode is actually slow! This is how they used to lie on the box specs: by using only a single number that is fast, while the rest are slow... and your tests only support the lie!


    Examples of how to test response times:
    Xbitlabs tests a range of 72 transitions spread out evenly; the also show the overshoot error rate for all those transitions:
    http://www.xbitlabs.com/articles/monitors/display/samsung-sa850_8.html#sect0
    TftCentral tests 30 transitions & the corresponding overshoot errors:
    http://www.tftcentral.co.uk/reviews/content/benq_bl3200pt.htm#gaming


    If you want, compare your tests of the BenQ BL3200 to tftcentral:
    http://www.tomshardware.com/reviews/benq-bl3200pt-qhd-monitor,3898-9.html
    The tftcentral tests show that the monitor has response time problems of 41ms for the 0-50 transition, but better numbers (6-10ms) otherwise. Your single number tests do not reveal any of these issues.
    0
  • KevinAr18
    Sorry for the aggressive tone in my previous comment. I should have taken more time to write it up nicely.

    This issue with response time testing has always been a problem on tomshardware (even before you ever started writing reviews here), however this is the first time I got a good chance to contact the person that writes the articles. I hope you will be able to look into the response time tests at some point. Sadly, there is literally only two sites on the internet (that I know of) that test response times correctly: xbitlabs & tftcentral. In fact, even tftcentral used to test response times wrong (for many years). xbitlabs was the original site I know of that began testing response times correctly.

    I apologize for not being able to write in more detail right now, but here's two helpful links:
    Response times:
    http://www.xbitlabs.com/articles/monitors/display/lcd-testmethods_5.html
    Overshoot errors:
    http://www.xbitlabs.com/articles/monitors/display/lcd-testmethods_6.html
    0
  • dovah-chan
    You can't really fully benchmark response times accurately as it varies amongst monitors such as the overall overclocking headroom varies by CPUs that are even of the same model.

    Also thanks for this article. It's n-not like I was one of the people who requested it. I'm just glad to see that we are listened to. >__<
    0
  • Drejeck
    I sense a disturbance in the force.
    Is some monitor being reviewed right now? Hope it's a G-sync near 300 euros
    0
  • Blazer1985
    Hi christian!
    I have a request :-)
    Would it be possible for you to compare some cheap chinese tv and monitors to the expensive western branded ones?
    I know that you usually get what you pay for but there might be a sweet spot somewhere in between and would be nice to know :-)
    Especially today that we are beginning the switch to 4k and the prices vary so much (ex. Haier 50" uhd @600€).
    Thanks!
    0
  • KevinAr18
    Anonymous said:
    You can't really fully benchmark response times accurately as it varies amongst monitors such as the overall overclocking headroom varies by CPUs that are even of the same model.

    The part I like the most about "benchmarking response times" is that even with minor variations between each monitor, the numbers are very useful in finding out if a particular model is any good. (Personally, I think that the variations between monitors of the same model may be so minor, that testing is still very useful, but I would need lots more data to say with more certainty :)).

    However, instead of me just making comments, I think you may appreciate a good explanation, since I failed to provide one before; sorry about that. :) So... here goes.


    To be clear, the problem that I was referring to with tomshardware's benchmark was basically this: tomshardware tests only one out of thousands of transitions, which may tell you nothing about the monitor's real response times!

    The best way to understand what I am referring to (and why it is so important) would be for me to try and explain the details.

    Consider these details about transitions & response times:

    1. An 8bit monitor uses 256 shades of grey to produce each color. 0 = black; 255 = brightest color (white); 22= very dark grey, etc...
    2. Response time is the measure of how long it takes to change from one shade to another. Say from 0 (black) to 156 (greyish).
    3. A little bit of math shows that if you have 256 shades, there are 65,280 possible transitions. [ (256*256) - 256 = 65,280 ]
    4. Each transition has a different response time; that means an 8bit monitor has over 65,000 different response times!
    5. Consider this example:

    source: http://www.xbitlabs.com/articles/monitors/display/zalman-zm-m240w_4.html#sect3
    Notice how the 255 to 0 transition (white to black) has a response time of ~2ms.
    However, look how many other transitions are maybe 14ms or more!

    Sidenote: overdrive:
    1. Long ago, they dicovered that they could increase response times by applying more voltage to certain transitions (or maybe less in some cases?? not 100% sure).
    2. This "overdrive" method can improve response times a lot (you want it on a monitor). However, if not tuned correctly, too much voltage may be applied and it may "overshoot" the correct color and create bad image problems that are sometimes worse than having slow response times.
    3. This means any response time testing must also test for overdrive errors (commonly called "RTC error").


    The problem with tomshardware (and most sites):
    Currently tomshardware tests only the 255 to 0 (white to black) or the 0 to 255 (black to white) response time and ignores all the other 65,000+. As you can see from the earlier picture, if you test only one transition and not the others, you may end up with a response time that is 100% useless and misleading.
    Also, tomshardware does not test for overdrive errors, which, in some cases, can sometimes be more important than response times

    The correct way to test response times:
    Obviously, you don't want to test all 65,000+ transitions. However, you can test a range of them. Xbitlabs tests 9x9 transitions or 72 transitions scattered evenly around ((9*9)-9 =72).
    This testing reveals some very interesting things. Consider these examples:

    http://www.xbitlabs.com/articles/monitors/display/20inch-6_15.html
    This monitor has a white to black transition of 2ms and a black to white transition of 6ms, but 55+% of the time it is really 16+ms! Tomshardware would have reported false (low) response times if they had tested this monitor using their current method.


    http://www.xbitlabs.com/articles/monitors/display/acer-gd245hq-lg-flatron-w2363d_9.html#sect0
    The response time on this one varys a lot, but you'd never know that without the tests done like on xbitlabs.


    http://www.xbitlabs.com/articles/monitors/display/20inch-4_23.html
    This picture is actually not response times but overdrive errors that I mentioned earlier. What this shows is how MVA panels often have MAJOR image problems when dealing with 0 to 96 and/or similar nearby transitions, but is good most other times. AKA, it means you'd see bad glowing on some dark scenes but a good picture the rest of the time. Note how tomshardware tests reveal none of this important data.
    0