[Info] AMOLED Power Consumption Tested and Explored

As I was reading an article on my phone one day, I wondered about AMOLED screens and their power consumption. These screens have been widespread since the original Galaxy S, and they’re also used exclusively on the Note and Tab S lines. AMOLED has also been used in Nokia, LG, Motorola, Oppo and now even Apple products, via Watch. Being that the screen component is by far the greatest battery drain on a device (around 80% during normal use), and battery woes are a problem that plagues us all, I set aside a few hours one day to objectively measure just how much effect it has. Here are the questions I’ll be attempting to answer:

  • Exactly how much additional battery power does increasing screen brightness consume on AMOLED?
  • How much less power does using a black background on AMOLED screen save, vs a white, or gray background?
  • What is the effect of specific Average Picture Level (APL) on power consumption?
  • What is the optimum brightness setting and color background to use, in day to day use?

An AMOLED Primer

AMOLED (Active Matrix Organic Light Emitting Diode) is a type of display technology that uses a matrix of individually-switched self-illuminating diodes that can be controlled and lit without a master backlight. Samsung markets theirs as Super AMOLED, which usually uses a ‘pentile’ matrix layout, or an S-Stripe layout. For the purpose of these tests, I will refer to Super AMOLED as just AMOLED. Read the wiki page for a quick primer. AMOLED screens are known for their very high levels of contrast, lack of backlight bleed, much faster refresh rates, lower power consumption and very wide color gamuts, compared to their traditional LCD counterparts, whilst retaining viewing angles and color accuracy, at least in a properly calibrated device. They are especially suited to consuming media, due to the vast color and contrast available, are usually thinner than LCDs and can be curved due to the nature of the lighting. OLED TV’s have long since been treasured, but low yields in large panels relegates them to relatively small devices for now.

The relevant information to us is that firstly, individual pixels can be controlled and lit without a single large backlight, such as in a traditional screen. Power consumption for a completely black diode is zero (powered off), whereas for an LCD screen, displaying a black pixel requires attempting to fully block the light being shone behind it – there’s usually at least a tiny bit of light leakage, especially towards the edges where the backlight(s) rest. On many AMOLED devices, displaying a pure black screen is usually indistinguishable from darkness.

The second relevant point is that the different colors of the AMOLED matrix are not evenly distributed. Some colors are found more often or in larger physical form than others. The reason is due to the efficiency of light production – green diodes, for example, are 12x more efficient at light production than blue diodes, and 1.8 times more efficient than red diodes, at producing the same amount of light (source). AMOLED manufacturers thus use Pixel Area Scaling (PAS) to compensate for this imbalance, often making the blue pixels larger. Just to add confusion, our eyes are sensitive to different colors at different light levels – blue at low light intensity, red at high intensity and yellow/green in between. Keep this in mind for later. 

s5amoled

Microscope View – Galaxy S5 AMOLED Panel. Notice two green diodes for every one red and blue diode. Source: Anandtech

AMOLED Downsides and Degradation

It’s not all good news, previous AMOLED screens tend to have slightly higher reflectance and poorer sunlight visibility. In addition, because of the uneven application of colored diodes, factory color calibration is more important than ever to correct imbalances. Some AMOLED models suffered from a blueish or greenish tinge. Roughly 25% of the devices I’ve owned were AMOLED devices, the Galaxy S, Galaxy Nexus, Note 2, Nexus 10 and Tab 7.7, all the way to the Galaxy S5 and Tab S 10.5 used in these tests. I’ve seen the mixture of calibration and pentile arrangements first hand, with the worst being the Galaxy Nexus, which is about where Samsung started to experiment with curved screens.

Earlier AMOLED generations, such as the S3/S4 (and the Nexus 6, which uses a panel similar to the S4), have reported burn in with white, or more accurately, the blue component of white colors. Blue diodes age dramatically quicker than the red or green diodes, they shrink over time, emitting less light and are susceptible to burn in if displaying the same image continuously (such as display models). This may or may not be an issue for you depending on your usage pattern and how long you plan to keep your device. In addition, lower resolution (< 350 ppi) screens make the pentile layout more visible, reducing the sharpness of elements.

On the topic of burn-in over time, the complexity occurs as there is not one backlight that ages, but millions of them, and they age at different rates. Note in the graph below blue diodes age roughly 20x as fast as green diodes, with reds lasting the longest. Similar results are presented in this Densitron presentation, which grades blue at roughly 2.5x worse than green. Before everybody panics and riots, the absolute numbers here are not important, just the relative ones. If you purchase an AMOLED device now, it’s not going to self-destruct in 700 hours of use. There are people still using Galaxy S1’s from more than 5 years ago without burn-in.

oledlife

 Source: UniversalDisplay

Fortunately, each generation of AMOLED has steadily addressed these issues. The Galaxy S5/S6 devices have an ultra-bright sunlight mode which somehow boosts light production from ~400 nits to 600+ nits, which is on par with the brightest LCD panels (iPhone / Xperia Z3). AMOLED used to be very inefficient on white displays, but two out of the top three devices in 2014 for battery life were AMOLED devices. Finally, power consumption at higher resolution appears to be have been addressed, as the S6 panel has been tested to consume less power than the S5 panel, despite having roughly twice as many pixels to light. The factory calibration of modern AMOLED devices is also leaps and bounds better than early panels, being rated as the best overall display on any smartphone by Displaymate, even since last year’s S5.

The Difficulty of Measuring Power Consumption

At first glance, one might think it’s fairly easy to measure instantaneous power consumption. But getting any kind of accuracy required for this type of testing is far more difficult. The available battery apps, including Ampere (which I’d highly recommend for testing USB cables), are only as good as the information which the device is getting from the hardware sensors. In the case of the Galaxy S5, the minimum granularity achievable was a -450mA power draw. To be able to use this devices, I measured the voltage drop over an extended time (usually 20-30 minutes), then averaged to smooth out the irregular nature of voltage measurement. The Galaxy Tab S 10.5 was able to measure a far more precise power draw, down to -15mA or so, but I still included one test from the S5 for thoroughness.

Why does these voltage discrepancies occur? Unfortunately, battery technology is not really that advanced, it hasn’t progressed that much in the last decade, nowhere near as quickly as the rest of our devices. The vast majority of batteries being used in phones are Lithium-Ion (with graphite anodes), which generally operate between 3.6V-4.2V, depending on charge level. To estimate battery capacity, instantaneous voltage readings are read by the phone, smoothed and measured against what the phone knows about the battery, to produce a % charge level. Overall, it works for day-to-day use, but for testing from minute-to-minute, the voltage reading tends to jump around because of the nature of batteries (source). We need to keep this in mind when testing.

voltagevstimeVoltage drop is generally stable between 3.6-4.0V for Li-ion, notice the thousands of small fluctuations. Graph source: Richtek 

 Testing Methodology

  • Turn off all background tasks / sync / apps.
  • Only test device between 3.6-4.0V range, as the voltage drop appears more linear in that range (source).
  • Since the brightness slider is not precise, brightness is instead set directly via editing the value (0-255) at “/sys/devices/mdp.0/qcom,mdss_fb_primary.190/leds/lcd-backlight/brightness” or “/sys/devices/platform/s5p-mipi-dsim.1/backlight/panel”, depending on the device. Note: Value of 0 is lower than 4% that the brightness slider achieves at it’s lowest.
  • Battery temperature stayed roughly between 26-30C (79-86F) for all tests.
  • All brightness values are provided in nits (otherwise known as candelas, lumens or cd/m^2).
  • Power draw is measured in either voltage drop per minute, or per interval. Relative value is more important, as it measures TOTAL device power draw, not display only.
  • For 100% APL tests, a pure white image is displayed. Various APLs are achieved with precisely measured grayscale images to achieve 10% APL granularity. A pure black image is used for 0% APL.
  • Device is left with screen on, at set brightness and APL, around 10-20 minutes during each test, for battery power draw to stabilise.
  • Connecting to the phone via adb over network, to avoid charging the phone via usb. Read battery values from:“/sys/class/power_supply/battery/uevent”. 

Unfortunately, that’s as accurate as I can achieve with this setup. Further accuracy can be gained to some degree by averaging over an entire battery charge, but that would take a prohibitively long time. Alternatively, measuring directly from the battery lines or somewhere on the power IC would be an option, but I will avoid taking apart these devices for now. Later, I will compare results with others who have completely similar tests.

Test 1: Effect of Screen Brightness on Power Consumption

The purpose of this test is two-fold – test how the actual light output of the phone varies depending on the brightness setting, and to test what effect the brightness setting has on power consumption. A pure white image (100% APL) is fixed on the screen, brightness is adjusted in 10% increments precisely and voltage measurements are taken in 10-20 minute intervals after each change.

brightnessvspower

Similar to LCD screens, we can see that power consumption increases exponentially towards the upper end of the scale. Above 80% brightness, it skyrockets to eke out that last incremental increase in brightness, a poor trade but a necessary one. The remainder of brightness consumption follows a reasonably linear trend. I also noted that the screen being completely switched off (sleep mode), resulted in an approximate -20mV power draw in total, from other components. It’s reasonable to presume that the CPU governor also behaves in a more conservative manner in sleep mode, but there’s still a 20x jump in power consumption as soon as the screen is switched on, even when displaying nothing.

Test 2: Effect of APL on Power Consumption

The purpose of this test is to examine the effect of different APLs (Average Picture Level) on power consumption, which is a good indicator on the average brightness content on the screen, not just peak brightness. TV shows / movies are usually around 40% APL, while mostly white webpages are around 80%. The jump from Android 4.4’s mostly-dark Holo theme (around 40%) to Android 5.X’s (80%) mostly-white Material theme also affects APL significantly. More detailed analysis of APL can be found at Anandtech. Note that a pure red, green or blue screen, at the same brightness as a pure white screen, will be 1/3 the APL on average, as only 1 out of 3 diode colors are being lit fully. It’s also important to note that brightness and APL are separate – a screen with exactly 50% pure black and 50% pure white halves, will be 50% APL (the average), but will measure a maximum peak brightness on the white component.

aplvspower6 APL intervals as tested on Galaxy S5, fixed 200 nit brightness. Orange is linear trendline

As we can see, power consumption increases nearly linearly with APL, even as screen brightness is a fixed value. A pure white display uses more than 4 times the power than a pure black display. In practical terms, a mostly white webpage uses about 65% more power than a mostly black movie/TV show or app, all else being equal. Also note that difference in scaling compared to brightness, there is no exponential rise at the top end, and there is still a significant power draw when the screen is displaying a 0% APL (equating to all pixels being off). We can presume this is due to the display controller and other components not being in sleep-mode.

  • Test 3: Effect of Turning Off Specific Diodes (Varying Color Presets, Fixed 200 nit brightness)

Changing the colour of individual pixels on an AMOLED screens greatly affects the power consumption, unlike LCDs. If the pixel only has to light a pure red, green or blue image, or some limited combination thereof, it’s less wasteful than lighting up pure white, which requires all three components to activate. To test this, I used CF.Lumen (by the legendary Chainfire). Unlike Twilight or Lux, it doesn’t simply apply an overlay to the screen, but actually adjusts the color value of each displayed pixel. No washed out blacks, and the ability to infinitely adjust color balance, even to extremes.

colorvspower

 

Now this is tremendously interesting, it confirms not only that the pure white screen takes more than twice the power of a pure black screen, but that displaying pure colors varies greatly. Blue is the most inefficient color, requiring nearly 30% more power to display than a pure green screen. Amber (which is a mixture of 100% Red and 75% Green, with Blue being disabled) is surprisingly more efficient than Red, but keep in mind there is some margin of error in these measurements. Salmon (100% Red, 50% Green, 50% Blue), is nearly as inefficient as pure blue, but still a a fair bit behind a pure white. Both of these are used commonly as alternatives to pure white, especially Amber during Sunset hours, to aid in sleep. Amber and Salmon are perfectly readable, whilst pure green, red and blue are very harsh.

Comparisons with Similar Tests

  • Anandtech ran a similar test, but in a more accurate fashion, measuring directly from the circuitry. The results can be found here – The difference between a minimum and maximum brightness, displaying a 100% APL, was 452mW – 1517mW, or approximately 3.35x power consumption increase from adjusting brightness, for the same image displayed, less than the 4.4x increase I measured in total power draw, but in the same ballpark.
  • Greenbot did a related test, using an app with a standard and a black theme, then measuring the average difference over 10 measurement periods. They found a 41% decrease in power consumption from using a black instead of a standard theme, whilst keeping the brightness locked to 50%. In my testing, I found a 65% decrease in power consumption from the a similar change in theme, also roughly the same.
  • Displaymate measured total device power draw between minimum and maximum brightness on the S5 (0.82w – 1.5w, or 82% increase) and the S6 (0.65 – 1.2w, or 84% increase) with a 50% APL. It’s also interesting to note that the S6 display is around 25% more efficient, despite having nearly 2x more pixels to display. I tested brightness adjustment at 100% APL, not 50%, so am unable to compare results, but included this for thoroughness.

Conclusions and Guidelines

With all of the above data, we can now form a few conclusions and practical guidelines for AMOLED screens. It’s important to note that AMOLED technology has progressed very rapidly, and it’s highly likely that it will do so in the foreseeable future. The S5 and Tab S used for testing is approximately on par with the Note 3 in display technology, which has already been superceded, but the principles should hold the same. I wrote this article primarily to satisfy my own curiosity, but I’ve left some guidelines below for those that want to translate it into day-to-day battery saving tips.

Findings of AMOLED screens:

  • Typically, the screen consumes more than 80-90% of total power consumption during normal hands-on use.
  • AMOLED screens at 100% APL (pure white screens) are more inefficient than LCD screens. For every other situation, they are more efficient, as they can be selectively lit.
  • At a fixed brightness (200 nits), displaying a pure white image requires nearly 2.2x more power than a pure black image.
  • A mostly white screen (80% APL, such as a typical webpage / Material theme), requires 49% more power than a mostly black page (movie / TV show / Holo theme).
  • The scaling of APL to power consumption is nearly exactly linear.
  • 100% brightness setting consumes more than 5.4x the power of 0% brightness setting, to produce 30x the brightness.
  • However, 100% brightness consumes 1.7x more power than 80% brightness, to produce only 1.2x more light.
  • Blue is the most inefficient color to display by far, followed by red then green, which are close.
  • Amber is on par with efficiency as pure red or green, but is far easier on the eyes.

Battery-Saving Guidelines for AMOLED screens:

  • Switch from a colorful wallpaper to a pure black wallpaper on your homescreen (which I measured on mine at 59% to 19% APL). Doing so will net up to 26% increase in screen-on battery life, at typical brightness.
  • Switch to black or AMOLED-friendly themes in apps which support it, I found around half my apps had black themes. Assuming typical APL levels, this will net up to 49% increase screen-on battery life, at typical brightness.
  • If you have root access, install and use CF.Lumen to apply Amber-type color schemes. Not only is it less harsh on the eyes (especially for reading for extended time), but it will net up to 36% increase in screen-on battery life, at typical brightness.
  • If you have an adjustable smart brightness control, try to decrease it about 20% across the board, this should still let things remain legible, but also net up to 23% increase in screen-on battery life, in typical scenarios. Better yet, use the lowest bearable brightness.
  • Just switching the phone on at all increases power consumption by more than 20x-40x, use Ambient Alerts (which only light a small part of the screen) or some other notification system (like Notification LEDs, which take nearly no power).
  • All the above increases are theoretical only, in optimum, controlled situations where only the display, and not CPU/GPU is being used. It’s highly unlikely that you’ll gain such large benefits in day-to-day use, but there will be noticeable benefits, probably 1/3-1/2 of the stated figures.
  • Additionally, these calculations are for the display only with no background tasks, your background syncing, apps, radios, GPS, will all consume more power. You’re on your own for screen-off time.

All feedback, comments and corrections welcomed.

Advertisements

10 thoughts on “[Info] AMOLED Power Consumption Tested and Explored

  1. I have a question with regards to OLED power consumption when producing white, or some combination of colors. Wouldn’t you be able to have each red, green, or blue pixel emit at a lower intensity than if you were producing a single color and then have their intensities combine? If that’s the case, why would producing white or other non-primary colors use more energy?

    1. Hi Ben, since the OLED only has R, G and B pixels (as see in the close up), in order to produce white they need to all be switched on and combined, hence the higher power draw when producing white then any other color. As for non-primary colors, since more than one type of pixel has to be switched on and combined, there is also a higher power draw than just producing one primary color by itself. Hopefully that clears it up.

      1. Thanks, I’m thinking of each R,G,B pixel as separate bulbs. I could theoretically light 3 dim bulbs each 1/3 as bright as another single bright bulb and have the brightness levels and power draw be the same in each scenario. If you wanted to display white at 60% brightness, could you combine each R,G,B at 20% brightness (20+20+20), or does it not work like that?

      2. I see what you mean, there is an additive effect, but it’s not 100% efficient. For example, see the ‘Pure Colors vs Power Consumption’ chart in the article. There is no yellow diode, and to generate yellow would require mixing two colors together, but the power consumption for yellow is not the full sum of them together. Also, anything which involves blue requires a large amount more power (including purple and white) as it’s more inefficient / larger diodes. At a guesstimate, displaying pure white requires about 40-50% of the power of lighting each color separately at full brightness all added together. I’m assuming the AMOLED engineers have fine-tuned all of these mixtures in the display controller.

  2. Can you clarify? In reference to the UniversalDisplay table of pixel lifetime you say that red lasts the longest but based on the lifetime hours yellow does? Am I reading this wrong? Why does yellow last the longest if it requires multiple pixels just to make the color?

    1. You are correct, yellow lasts longer than red, but there are no yellow OLEDs on our phones/tablets, thus red is the longest lasting on our devices.

  3. Pingback: AMOLED | AxelBlog
  4. Very helpful article. I was surprised by how much extra battery life I obtained from changing S6 homescreen background to black, so I searched to your article looking to find out what single color would preserve most of the benefit while looking a lttle nicer. Looks like amber is a good bet. Thanks.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s