mux' blog

Looking for the Dutch version?

In this blog series on display power consumption:
Part 1 (Anatomy) - Part 2 (Ideal display) - Part 3 (You are here)

Previously in this blog series I explained the operating principle of displays, and then explained all the theory behind display technology from a power consumption perspective. Today I present to you my research into actual LCD power consumption. I compiled measurement data from two major websites and tried to find out if there are some interesting patterns.



This blog series accompanies the blog series about my newest computer build, an Ivy Bridge desktop quad-core all-in-one with IPS display consuming only 20W when idle. If you like this series or have found the information in my blogs useful, please consider donating to me (donation link at the bottom of each blog) to help make my newest build as awesome as possible.

0. Index

• 1. Correlative study of LCD monitor power usage
  • 1.1 Hardware.info is bad at measuring
  • 1.2 So, what original research did I do?
• 2. The measurements
  • 2.1 Interesting correlations
  • 2.2 Which brand and model to choose?
• 3. Conclusion

1. Correlative study of LCD monitor power usage

Most of the text in this blog is basically what is known in scientific terms as original research: stuff I found out by myself. This is necessary to some degree because to my knowledge nobody else is doing this kind of research. So what original research have I done? To put LCD power consumption into perspective, I have compiled a list of 292 PC display test results which contained power consumption and white point brightness figures. Here we go:



I am not yet going to tell you what all the numbers mean. Just pretend you are really impressed by the research I have done. First I have to talk with you about the wonders of benchmarking. You will never find two sites that have the same results, obviously. Computer benchmarking specifically is very bad in this respect. You will rarely find two sites that agree within even the most lenient of error margins, for instance 10%. And worse still, there is usually absolutely no way to explain or verify why results between benchmarkers differ. So, why the warning? Because of this:

1.1 Hardware.info is bad at measuring

This is a graphical representation of the differences between my two resources: Prad.de and Hardware.info. Those are the only two sites I know of that have a comprehensive and mutually overlapping database of white point and power consumption measurements. Every data point (there are 36 in total) is a display that both Hardware.info and Prad.de have tested. On the horizontal axis is the relative difference in power consumption, on the vertical axis is the relative difference in maximum display brightness.

Just for reference: 0.1 on this chart means the benchmarking results are within 10% of each other, which I would consider quite a large margin of error. Out of the 36 displays, only 14 have test results within 10% of each other. The two sites very roughly agree about 39% of the time. All other measurement data is much more different, with a couple of extravagant outliers:

• Dell U2312HM, with a 54% brightness and 37% power consumption discrepancy.
• NEC PA241W, with a 51% power consumption discrepancy, but almost no difference in measured max. brightness
• Dell U2410, with a 69% brightness and 71% power consumption difference
• Fujitsu P27T-6 IPS, with an 84% brightness and 70% power consumption difference

There is some excuse for display brightness variations between reviewers, but there is absolutely no excuse for such gigantic discrepancies in power consumption. Because of the way backlight inverters work (they have a fairly precise current setting) they will rarely vary more than a couple of percent. Furthermore, other sources of power consumption are probably insignificant in these comparisons: the displays are all tested with a white test screen (so there should be no difference in panel power consumption). The only variation that might be left is the connection used - VGA, DVI, HDMI or DP - but that part of the electronics in screens usually doesn't use more than about 1W, so it should be negligible.

However, there is a very clear trend in the differences: most displays with a (much) higher power consumption at Prad.de also seem to have a much higher reported maximum brightness. There seems to be almost a linear trend - excluding the NEC. This leads me to suspect that Hardware.info does not properly test displays. Instead of reporting actual maximum brightness (like their data seems to imply - it literally says 'max. brightness'), they test the maximum brightness at stock brightness settings. Why do I think this? Almost all modern displays are set to maximum (100%) brightness at the factory, however some displays are shipped with lower settings. Many Dell displays have a default setting of 75% for instance. Let's see if the numbers match up:

• Dell U2312HM ships, like almost all Dell displays, with a 75% brightness setting. This matches the reported figures in Hardware.info almost exactly, as well as the power consumption (note that in the Prad review, the brightness setting is not linear with actual screen brightness, which accounts for the fact that power consumption is also not linear in these settings).
• NEC PA241W ships with the display set on 'Eco', which limits the brightness to the 35% setting - almost exactly matching the 65.1W value Hardware.info reports. This is done because the backlight has a very high maximum brightness.
• Like the NEC, the Dell U2410 is a display with very high maximum brightness, so it ships with a brightness setting of about 250 cd/m².
• Again, the Fujitsu display, this time Prad.de explicitly noted in the power consumption table ships with 35% eco mode engaged.

This is just a small selection of the linear discrepancies, but the general pattern continues for many more displays. I am fairly sure that Hardware.info has produced a lot of erroneous data and essentially cannot be trusted for their display reviews. The most compelling evidence is the large proportion of positive linear outliers in the test data, but their test data seems to be very rushed in other areas as well. This should come as no surprise, as this website has churned out more than 500 display testing reports over the course of 1.5 years. This has largely been under the supervision of one person. Prad.de has a small team of people who have produced just over 100 extended reviews over the course of 10 years. They have definitely had more man-hours with each piece of hardware.

1.2 So, what original research did I do?

Now back to what this blog is actually about: power consumption. As I have mentioned in the past, display power consumption is defined by essentially four factors:

• The backlight - this is the main consumer
• The panel itself - 2nd biggest
• The microelectronics - small contribution
• The power supply, albeit built-in or external

Unfortunately, there is no way for me to actually measure power consumption differences between displays. But I do want to explain power consumption numbers in a meaningful way. Through the knowledge of displays developed in the last two blogs (okay, and a metric crap-ton of practical experience) I have developed the following graph as a visual aid in our journey to understanding:



This is a model of display power consumption, relating the real world power consumption to what is going on inside a display. Notice that I chose the vertical and horizontal axes to represent something you might come across in a display review like those on Hardware.info and Prad.de. Okay, so what does this all mean?

First, there is the symbols 'X' and 'Y' on the vertical axis. These numbers are the minimum and maximum power consumption numbers you may find in a review. The minimum power consumption is usually measured at minimum brightness, which is represented by the symbol 'A' on the horizontal axis. Likewise, maximum power consumption occurs at maximum brightness 'B'.

However, power consumption is not just a function of the backlight. For one, the panel itself consumes quite some power as well. When displaying a completely white image on an IPS or VA matrix, or when displaying a completely black image on a TN display, the power consumption of the panel is higher than when displaying the opposite images. This power consumption is irrespective of display brightness, so you can visualize this as the entire power consumption curve shifting up and down as pictured.

Note that even at brightness A, the backlight is not completely off and is thus still using some power. We can extend the solid red line to a fictitious zero brightness to arrive at the power consumption without the influence of the backlight and panel. This gives us an indication of the power consumption of the microelectronics (and power supply overhead).

Talking about the power supply: the red line is curved because power supply efficiency usually increases towards higher output power. A small increase in DC power consumption at the low end results in a larger increase in AC power consumption, whereas at the high end this relationship is much more linear. However, in the highest regions power supply efficiency decreases again, causing the same behaviour as in the low end.

The numbers A, B, X and Y are measurements you can find in reviews. With these numbers you can work out, using the model I just explained, roughly how much the backlight uses, how much the panel uses, etc.. So what can we do with these kinds of results?

Well, all kinds of interesting things. For instance, it is possible to find out who makes the most efficient panels, backlights, microelectronics and power supplies in displays. And that is what I am extremely interested in - all this information, all this research eventually has its application in either advising people which displays to buy or to tell display manufacturers and assemblers how best to improve display power consumption. So, let me get to the measurements!

2. The measurements

Finally, the meat of the post. What I have done is compile a list of PC displays, their maximum brightness and associated power consumption when displaying a white image. Ideally, I would like to compile many more measurements but this is as much as I could find on at least two websites with overlapping data, and research rule #1 is: always verify your results in some manner. If not with your own measurements, do it with multiple sources. As mentioned before, even these sources do not agree, but their disagreement can be explained which is good enough for now. The results can be found in raw format in this excel document. I highly encourage anybody to do interesting analysis on the numbers, and by all means point out errors I made!

Okay, so we are missing crucial data in order to reconstruct the model. That is no problem; we could start by simply comparing power consumption and see if there are interesting patterns. However, the data I have compiled consists of wildly different displays: some are 22 inches in diameter, others 27 or even 30. Some have a maximum brightness of 200 cd/m², others go as high as 450. Some have CCFL backlights, others LED. And of course the different panels. We need to standardize the measurements somehow. This is where MAWTDEN comes in.

MAWTDEN is short for Mux' Arbitrary Weakly Tested Display Efficiency Number. It's a unit I thought up just for this blog and it is a very coarse measure of a display's efficiency. The MAWTDEN is calculated as follows: calculate the surface area A of the display. Multiply that by the maximum brightness B and divide it by the associated power consumption P: A*B/P. For instance, my LG IPS231p has a maximum brightness of 240 cd/m² and a surface area of 0.146 square meters. Its power consumption when displaying a white image at this brightness is 33.4 W. So the MAWTDEN of the IPS231p is 0.146*240/33.4=1.05 (in units of candela per Watt, cd/W). What does this figure of 1.05 mean?

A higher MAWTDEN, for example 2, means the display is more efficient - it can display a brighter image while using less power. A lower MAWTDEN means the converse. As it turns out, MAWTDEN is a very nice figure because a figure of 1 is very close to 'average' - actually, the average of all displays is 1.29. The best displays can exceed a MAWTDEN of 2.5, while the worst displays barely make it over 0.5.

2.1 Interesting correlations

Now let us dive into a bit of statistics. How large are differences between types of displays, manufacturers of displays and even different generations of displays? One of the first things that comes to mind is the recent revolution in using LED backlights instead of CCFLs. How much has that impacted display power consumption? And what about MAWTDEN?



This is a scatter plot of TN displays. The red dots are CCFL-backlit displays and the blue dots are LED-backlit displays. As you can see, and as I have pointed out with the rather blatant arrow saying 'better', LEDs have indeed reduced overall display power consumption and increased MAWTDEN. The entire 'cloud' of data points has shifted in the 'better' direction. But the reason I made this a scatter plot and not just two bars saying the average of each class of displays is the fact that there are some very interesting outliers.

Look at the highest two red dots. These two displays have a very high MAWTDEN - these are very efficient displays, yet they use CCFL backlights! Which displays are we looking at? They are two slightly different measurement results for the Iiyama E2710HDS-1. How did this display become so efficient? Let us look at the detailed measurements and try to fit it into our model, shall we?


Test results as retrieved from Hardware.info, 6th June 2012



Test results as retrieved from Prad.de, 6th June 2012

Two things immediately catch my eye: the very high maximum brightness and the very small influence that the panel seems to have. The difference between a completely white and completely black screen is just 0.8W. For a 27 inch display, that is extremely low. We can get a clue as to why this is so low by looking at the response times for this display: gray to gray switching times are fairly slow. As it turns out, this panel does not use a technique called response time compensation or RTC - which is a way to improve display response times. RTC is mainly used as a marketing tool to sell displays with very low '2ms' response times, but in practice this technique just wastes lots of power and does not really help much. Other factors are much more important in determining response times.

However, just the lack of RTC does not fully explain why this display is so efficient. The second clue comes from Prad.de's table which lists power consumption as a function of brightness. At the lowest brightness setting of 34 cd/m², power consumption is 11.7W while at the highest setting of 424 cd/m² it is 44.6W. That means that the display electronics actually are not that efficient - they use about 10W of idle power, but the strength of this display is that its backlight is very strong and at maximum brightness more than 75% of the power goes into the backlight instead of going 'to waste' into non-light-producing electronics. Bottom line: displays with a high maximum brightness score well on MAWTDEN. Can we get statistic proof for this?



This plot doesn't really seem to give us much to go at. There is a slight hint of a linear trend: at 200cd/m2 most displays are in the 0.6-1.6 MAWTDEN range, which gradually increases towards 350cd/m2. But it's not very clear still. I can shed light on the situation however:



If we segregate CCFL and LED backlights there is an extremely clear trend visible. CCFL displays indeed have a very strong linear relationship between maximum brightness and efficiency. This is logical: their backlights are less efficient, so they represent a larger proportion of the overall power consumption of the display as brightness increases. LED backlights have a fuzzier relationship, but it is still there.

I have also drawn trend lines in the graph. The actual trend lines do not say much (they do not intersect (0,0), which is crazy because at a maximum brightness of 0 cd/m2 the MAWTDEN is necessarily 0). But the steeper slope of the LED display efficiency as a function of maximum brightness is striking. I do not yet have a complete explanation for this, and I expect the main reason for this higher slope is the fact that LED backlights represent a lower proportion of display power consumption.

Okay, so we have established that a brighter display makes for a more efficient display (more efficient at producing light, that is). Also, LEDs are definitely more efficient than CCFLs. How about panel types?



Now this is an interesting graph. It is not much different from the first scatter plot I showed you, but now I have included non-TN data points as well. Again, the red dots are the CCFL TN displays, whereas the blue dots are TN LED displays. The green dots are new: they are IPS and VA-based displays. The green dots seem to exist in two separate regions: a lot of them sit somewhere between CCFL and LED backlit TN panels, and another bunch are way off in the distance with very low efficiency and screaming high power consumption. How extremely interesting.

What you are seeing here is the revolution called e-IPS and MVA. IPS and VA-based displays, especially S-IPS and S-PVA back in the day, traditionally required loads and loads of backlight to even be visible. This was caused both by the pixels having very small aperture ratios (a lot of wires and transistors blocking light) and the panel technology itself being less transmissive than TN. Remember those IPS displays on IBM Thinkpads that were barely readable at maximum brightness?

First AU Optronics and then LG.Philips Display developed A-MVA and e-IPS to address this problem. This was especially important as this coincided with the use of white LEDs as backlights, which were very efficient but also lacked the sheer amount of light output that CCFLs managed to produce. And that is exactly what you are seeing in this chart: the giant chasm between old S-IPS and S-PVA technology and newer A-MVA and e-IPS panels with LED backlights. Marvellous.

But there is more. Most of the 'efficient' IPS/VA panels you see in this chart are e-IPS, and that has a distinct disadvantage with the testing method used to generate these numbers. Remember that TN panels are 'normally white' while IPS is 'normally black'? Yeah, compared to the TN panels these e-IPS panels are at a disadvantage because they are being tested displaying white, which uses more panel power than TN displaying white. So in reality, those e-IPS panels may very well be shifted a bit further to the upper left to almost coincide with LED TN!

2.2 Which brand and model to choose?

Okay, last bit of analysis for today. What brands perform best and - possibly - why? Let me begin with the coarsest of analyses. I have compiled the average MAWTDEN for each display manufacturer who has more than 10 displays represented in my tables.



With a comfortable lead, Philips and Iiyama make the most power efficient displays. This is completely deserved - there are no tricks involved here, they are simply the best. Throughout LCD history Iiyama has always made very attractive displays: good value for money, reasonable performance and build quality, good power efficiency. Philips is mostly the same, but they are generally harder to find in stores and a bit more expensive. Unfortunately, this is where the fair comparison stops.

Most other display manufacturers on this list have some reason why the numbers are a bit distorted. First of all, the obvious: NEC, Eizo, Fujitsu and Dell have the lowest scores and this is entirely because they have a significant proportion of old-tech IPS and VA displays in the comparison. In the same vein, Dell is the 'best' of the bunch because they are the only company with the more modern LED-backlit e-IPS-screens also included, whereas NEC only has old style displays. Lastly, people who buy an Eizo really should not care about power consumption as they have very different unique selling points.

But then there is Asus. They have a very, very deserved last place among mostly TN-manufacturers. Asus consistently delivers poor power consumption figures, even for their so-called 'eco' and 'green' displays. This is very unfortunate because they seem to have put a lot of marketing behind their products. Like their 'EPU'-outfitted eco-conscious motherboards however, this is really just marketing and has no relation to real-world performance. I guess this serves as a reminder never to believe marketing.

Anyway, there are better ways to visualize the display manufacturers. Let's plot the manufacturers on that scatter plot and see who's who:



That is much better. Very viewable. Let us first filter out all the mediocre results and just show the best and the worst



Now it is clear why NEC, Dell, Eizo and Fujitsu have relatively low scores. A good proportion of their displays are higher-power models for applications that require very strict color accuracy. Fujitsu does have a couple of TN displays and they have fairly good scores, whereas Dell clearly has quite a presence in the higher MAWTDEN range with their e-IPS and 'professional' line TN offerings (e.g. Dell P2210). Conversely, Philips and Iiyama almost exclusively produce sub-40W displays with high power efficiency. The best couple of displays even on the previous scatter plot are Philips and Iiyama offerings. Now let's look at the middle of the bunch.



There is honestly not that much to say. The main reason why HP and LG seem to score a bit lower than Samsung is the fact that they have a few older displays with high power consumption. Asus cannot be saved; they have a lot of inefficient TN displays even in this comparison, which includes a couple of LED-backlit offerings.

3. Conclusion

This is all the statistical analysis I will do today. I could go on for another couple of blogs talking about the wonders of practical display power consumption, but that is no use. I have not begun to scratch the surface - for instance I have only barely touched on constructing a breakdown of power consumption into panel, microelectronics, backlight and power supply. I leave that for you to explore.

The conclusions we can draw from this analysis? If you are in the market for an efficient display and do not mind bad viewing angles (i.e. TN displays), Iiyama and Philips are your friends. Whatever you do, do not buy an Asus display if power consumption is high on your list of priorities. Furthermore, do not trust reviews or marketing. Lastly, I hope you have learned how to interpret power consumption and brightness data.

Thanks to my girlfriend, my mother, Devilly, Infant, pientertje, sebastius, Snowmiss and TheMOD for proof-reading

---

No PayPal? Message me or e-mail me