Well, it turns out the poor results are due to brightness throttling on white or near white content, which will vary depending on the power saving options that are activated. In Chrome and in the Stock Browser, this is further aggravated by the undefeatable activation of the Power Saving mode.
The measurements below were conducted with an i1 Pro 2 spectrophotometer and a profiled i1 Display Pro (for dark readings).
Maximum Brightness (Adapt Display/Dynamic/Standard Mode)
- Even without any Power Option engaged, white levels on white or near white content will be 15% to 20% below peak brightness (290cd/m² from 355cd/m²)
- Auto-Tone will clip whites further above 60% Average Picture Level
- Power Saving reduces brightness by 25%
- In Chrome and in the stock browser the Power Saving mode is activated as a default (surely in an attempt to fool battery tests into overestimating number of browsing hours on battery) - and cannot be defeated. It is therefore recommended to switch to another browser (e.g. AOSP) for better outdoor viewing.
The graph below outlines the brightness throttling mechanism in Dynamic/Standard/Auto Mode at maximum brighness.
As we can see, web browsing is the activity that will take the most serious hit, seeing as web browsing average picture levels typically falls within the 70% to 90% APL range. Although the throttling is to conserve battery life, this a bit sneaky from Samsung, as this tricks reviewers into overestimating battery life and consumers into believing they have the best of both worlds (brightness and battery life). There is a choice that can and has to be made.
Auto-Brightness (Adapt Display/Dynamic/Standard Mode)
Using auto-brightness instead of maximum brightness will actually enable a boost mode, which will activate under prolonged exposure to extremely bright light (I had to use my HTC One's flash at close range to simulate this).
- The boost mode offers a 35% brightness gain, however this gain starts to plummet at around 60% APL, which means that the gain for web browsing and Google Maps will only be 5 to 15%
- The peak brightness under prolonged sun exposure at 1% APL level is 478cd/m² (displaymate measured this at 475cd/m²). whereas the maximum brightness for pure white ends up at the unboosted level of 383cd/m²
- Without Boost Mode, the brightness curve in a birghtly lit environment is exactly the same as the maximum brightness captured in my earlier graph.
- The auto brightness -5/+5 radio button has no impact on the maximum brightness achievable - it will go full throttle if it has to. Only brightness at lower ambient light levels is impacted by this fine-tuning...
- There is no Power Saving mode override in auto-brightness, and those browsers even benefit from the Boost Mode.
Just to put the graphs into perspective, I am showing the maximum reachable brightness for various Android applications:
Android Menu: 5% APL | Maximum Brightness: 465cd/m²
Android Menu with radio buttons: 10% APL | Maximum Brightness: 455cd/m²
OLED fiendly web site: 15% APL | Maximum Brightness: 445cd/m²
Random Gallery Picture: 40% APL | Maximum Brightness: 410cd/m²
Google Play: 60% APL | Maximum Brightness: 395cd/m²
Anandtech with zoom on picture: 70% APL | Maximum Brightness: 350cd/m²
Google Maps: 75% APL | Maximum Brightness: 320cd/m²
Whatsapp: 80% APL | Maximum Brightness: 305cd/m²
Android Contact List: 85% APL | Maximum Brightness: 295cd/m²
Google Results page: 90% APL | Maximum Brightness: 290cd/m²
Google: 95% APL | Maximum Brightness: 285cd/m²
Greyscale and Color Gamut (Film Mode)
- RGB balance has a discernible green push above 50% brightness (less exposed in Anantech's review because Brian measures at 50% brightness rather than maximum)
- Color Space is oversaturated even in film mode (consistent with Anantech's and Displaymate's findings)
- Gamma will also get worse on content that has a high proportion of bright content, with brightness compression happening at the top end of the spectrum. For videos, it is therefore recommended to use 50% of maximum brightness or auto-brightness
White point color Temperature:
- 6410cd/m² in Film/sRGB Mode - again, this is at maximum brightness, the green dominance will be different at different brightness levels (as can be seen on the curves below)
- 6940cd/m² in Standard Mode, Dynamic Mode and Auto-Adjust Mode (I personally prefer that color temperature because it is more consistent across the whole scale, so the picture looks more harmonious)
Gamma is improved versus previous iterations but there is still a slight black crush and white crush happening at the extremities (nothing too major). The white crush will not be observed with smaller test patterns but will be worse for bright content! The black crush will be worse with Auto Tone on (to conserve battery, the transition into blacks is quicker).
Color gamut as expected is oversaturated. But what is more surprising is that it is oversaturated even in Film mode, although to a smaller extent. I went back to Brian's review, and this was also what he found. I believe two other reviews had different findings - but I may remember this incorrectly
Measurements in Film Mode
Color Space (Adapt Screen/Dynamic/Standard Mode)
Viewing Angles (Film Mode)
- Viewed directly, the display exhibits a green push in spite of near perfect color temperature
- Viewing at an angle reduces green and red and increases blue
- The most neutral white balances (Delta E of 2.8) can be obtained by viewing the screen at a 15° angle (you can do the test and notice how the green push disappears!)
- The color temperature at this angle is further away from the 6500K standard than for direct viewing but it is still more neutral (shows the importance of RGB balance)
The following charts shows the RGB balance, correlated color temperature and Delta E for viewing angles of 0° to 40°.
This is an important lesson: sometimes it is preferrable to calibrate at a somewhat higher color temperature to improve the RGB balance (generally a blue push is less perceptible than a green push). However in this case, Samsung's calibration was most certainly to improve the overall brightness of the screen (green has a stronger luminance than red or blue).
Throughts and Recommendation to Devs for Kernel Calibration
Essentially, the adaptative brightness without power saving options is to fool regular gamma testing and make it look flat even though it isn't - again to conserve battery!
Using Average Picture Level patterns, to keep brightness constant across the greyscale, we can clearly see that the gamma is too high (2.4).
By reducing brightness as IRE levels increase, the display conceals the fact that brightness increases too slow across the full IRE spectrium - because the final brightness ends up being lower, regular test patterns will say that brightness at each IRE level is where it should be when in fact it lags behind. It only catches up with target brightness at 100 IRE because brightness is being throttled more with each IRE level.
For example, at 50 IRE, the theorectical brightness should be 21.46% of white brightness, except the white brightness at 50% IRE is higher vs. 100% IRE. What is important is not the 100% IRE brightness, though. because it is throttled and has no impact on intra-picture gamma.
What is important is the 50% IRE brightness, which is higher. So while regular test patterns lead you to believe brightness is where it should be and gamma is 2.2, in fact it is lagging behind for the full brightness spectrum.
So in conclusion:
- We need to recalibrate gamma on this display by using APL test patterns - they are the only ones that should ever be used on displays with adaptative brightness where brightness is a moving target.
- We need to deactivate the adaptative mechanism (which would have the secondary effect of improving overall brightness)
But as long at the adaptative mechanism exists, we will never be able to have perfect gamma at all APL levels. If we choose to have perfect gamma at 50% APL, we will have brightness below target for low APL levels (since the 50% APL brightness will be lower than at lower APL) and it will be ahead of target for higher APL levels (since the 50% APL brightness will be higher than at higher APL levels).
So in essence, because of the adaptative mechanism, we can never have a perfect intra-picture gamma that works equally well at all APL levels.
The standard error will be smaller at lower brightness levels because the peak brightness range will be more narrow. But in boost mode the max brighness will be 70%higher than minimum brightness, so the standard gamma error at low APL and High APL levels will be higher, so black crush will be quite big for low APL and white crush will be quite big at high APL.
In a nutshell, if we want perfect gamma, we will have to acccept lower intra-picture contrast at lower APL and equalize brightness on the maximum achievable non-thottling 100% brightness. Since the display is very linear, this should yield us perfect gamma and RGB balance for all brightness levels that the user chooses between 0 and max user-selected brightness.
On plasma displays, throttling typically only happens above a certain brightness level. and it is enough to stay within that comfort range But the fact that brightness is being modulated for different APL levels on the S4 makes a good calibration that works at all APL levels impossible...
I can't believe how imaginative Samsung is when it comes to fooling reviewers. The display has potential, I just know that calibrating this thing with undefeated adaptative brightness will not yield perfect results - unfortunately.
The most pressing issue at the moment is the green push though and oversaturated colors, though. I am pretty sure that reducing color saturation will also impact greyscales and brightness on this display...A lot of fun coming our way