Why hi-rez displays (like the iPad 3) consume more power and generate more heat
After years of designing custom display systems based around highly customized FPGA-based real time image processors we learned a very interesting lesson: Power consumption and image processor cooling requirements are a function of image activity and display resolution. And the effect can be really dramatic. This is particularly true when going from static UI-type images to moving video.
Why does this happen? Modern electronic circuits consume power when they switch states. When a flip-flop inside a chip goes from 1 to 0 (or 0 to 1) power is required to make that change happen. A flip-flop is made out of a number of transistors. The effect I am referring to actually happens at the transistor level.
A typical display processor hardware pipeline has millions of transistors performing different functions. From wide polyphase FIR filters for scaling to complex de-interlacing hardware (which also uses polyphase FIR filters).
While the interface to the display may only be 18, 24 or 30 bits wide, the image processing pipeline can easily be 64 or 128 bits wide within the processor chip. This can easily result in hundreds of billions of state transitions per second within the chip.
Using the iPad as an example:
2048 x 1536 pixels
24 bits per pixel (assumption)
60 frames per second refresh
= 4.5E9 potential transitions per second
I say "potential" because the number of transitions are a function of the images being displayed.
When displaying relatively static and, more importantly, relatively uniform content, the transistors don't need to switch as often. On the other hand, if the content has many transitions, lots of switching takes place. And power consumption (and therefore device heating) can increase dramatically.
One example of such patterns that I remember testing probably ten years ago was a pattern where even pixels are white and odd pixels are black. In other words, as many full pixel transitions as can be supported by the interface. I remember burning my finger when I touched the FPGA. This, BTW, was an accidental discovery while trying to debug a hardware problem that only occurred --you guessed it-- when images where noisy and busy.
Video is the perfect spoiler here. And, if the video is noisy and of bad quality it can be even worst. With video you can have almost random switching patterns covering most pixels on the display. I also remember that helicopter-in-flight footage (camera aimed at the ground below) seemed to be particularly nasty because every single pixel is moving and there's potential for vibration and noise that simply makes everything jiggle.
I'm not sure if the thermal problems being brought-up by new iPad owners have anything to do with this, but hearing one of these reports on TV brought back memories of burning my finger on an FPGA when I accidentally discovered this little known side-effect of resolution. From iPad 2 to 3 the screen got four times as many pixels. And so, there's a potential for four times greater power consumption (by the display subsystem) and four times greater thermal requirements. I'm not saying that this is what's going on. I just thought that HN readers might find this interesting. Prompted by an offline discussion I thought it'd be a good idea to talk about the other effects. I focused my post around the power requirements of the image processor circuitry. However, there are two additional components to an LCD display system: Display logic and Backlight power. Display logic is, as the name implies, all of the circuitry required to drive the pixels. This is everything from the data connector on the raw LCD panel up to and including the TFT transistors on the glass itself and the transparent interconnects (also on the glass). The on-glass circuitry has complex parasitics and capacitive loading that also causes a fluctuation of drive current based on the randomness of the images. Here's sample data from a 24 inch 1920 x 1200 LCD: In very (very!) rough terms this current is also a function of display resolution: Twice as many pixels will demand double the current. Going back to the iPad 2 vs 3 example, the the new iPad has exactly four times the pixels of the iPad 2 display, therefore, it should require four times more display logic current than the older model under all uses cases. The backlight may or may not demand more current as display resolution increases. Comparisons of displays of equal physical size but vastly different resolutions do not reveal a this effect to be an absolute rule. A lot depends on the design of the panel and the internal optics. While you are correct, there is another aspect to power in displays: The most important component of LCD power is the back light that illuminates the display from behind. The overall transmission of LCD is only about 5% or so, which means that the back light behind is as much as twenty times brighter than what the user sees. Increasing the pixel density implies that the pixels are smaller. Which means that the circuitry takes a larger fraction of the pixel [1]. Which means that more of the light is blocked. Which means that the back light now needs to be still brighter to compensate. As a result, iPad 3 back light uses 2.5 times more power than iPad 2 back light [2]. I am interested in knowing more about the logic and interface power in the way you have described. Please let me know if it is possible to get in touch. [1] http://www.extremetech.com/computing/122725-what-the-ipad-3s... You are right, of course, however, I have built quite elaborate LED-based backlight systems and have metrics on this too. A few years ago we did a custom system with a 1,000 Watt (yes, one thousand) custom liquid-cooled RGB LED backlight. Performance depends on many things, from aperture ratios to the spectral transmission characteristics of the filters as well as how they match the emission spectra of the backlight elements. With regards to getting in touch, email me at martin_05 the domain is rocketmail and TLD is .com Interesting! Thank you for sharing! You are welcome.
As you can see, a pattern with maximal excursions in the data causes the display logic current (and therefore, power) to nearly double. 24 inch, 1920 x 1200
Idd for full black screen = 1,700mA
Idd for a black/white dot pattern = 3,050mA