Actually looked at the differences between a website in dark and light mode ?

Because it's not just the background color that is made darker... the text too...

So people crank up the brightness again to be able to read what it says...

This isn't really the part of it that makes them idiots[1]. Failing to clarify this is looking exclusively at LCD, at least on the blog summary is the real failing. I suspect people use dark mode for a variety of reasons, maybe they prefer the look, maybe it's more comfortable to read in a dark environment. The point about the text being made darker too is not necessarily really an issue, part of their takeaway is the palettes chosen will probably affect people's response, but this is how people respond to typical dark modes.

The big omission is that on OLED the situation will be completely different. Let's assume all people are doing is turning the brightness up so the text brightness is the same level as they would have light mode background, which I think is what you're suggesting, but now we're already making assumptions about what people's target level is, and that one is probably not a given. (Eyesight isn't linear, why should white on black require exactly the same levels to read clearly as black on white? You'd think the lighting environment might also play a role, although they did test that and no effect shows up in their sample. Science does include testing things that seem obvious, because they're not always true.). Anyway, if we're assuming people will adjust brightness so the peak displayed brightness is the same regardless of mode then the power use will necessarily be less than in light mode, because the sum brightness across the screen is lower and on OLED that's what affects the power demand. Not looked at.

[1] I don't entirely think they're idiots, but the way they've chosen to summarise in the blog post is actually misleading and possibly mildly harmful (since it could persuade people trying to save power on OLED to use bright modes), while they could have done the additional work on OLED. Conference papers don't usually get peer review rounds of revision, often just include or reject, which means while the work as it is is fine they probably haven't been asked to fill out the gaps they would for a journal article.

Shocker! Dark Mode not efficient on LCD screens which use a different technology to OLED.

Yup, in my experience Dark Mode on LCD is indeed difficult to read, so brightness needs to be cranked up. Better to use a more traditional theme. Have been doing this for ages.

For OLED the reverse is true. For OLED screens and power consumption this is an interesting watch:

https://m.youtube.com/watch?v=N_6sPd0Jd3g

I love the aesthetic of dark mode. It often looks much better than its light mode equivalent.

I suspect that more time and care might be being taken on colour choices in dark modes - they're not an inverse image, after all.

But I stopped using them because I realised I was spending more time with the screen, which is - ironically - worse for my eyes. So I switched back to light mode, and take breaks more regularly.

From the actual conference paper:

"Further investigation is also needed to determine whether the observed rebound effect applies to devices with OLED displays, and to quantify the energy trade-off"

I'll bet not.

This is really not ground breaking, although maybe most people don't realise. They tested on an LCD laptop screen (2017 MacBook Pro), that's where the graph on the blog comes from. As they address in the paper, but nowhere in the blog, LCD backlights are set for overall brightness, pixel elements are subtractive, and therefore power consumption is fairly independent of the displayed image, while LED/OLED pixels are self illuminated and so larger black areas reduce power consumption.

It is actually interesting to show that on LCD devices people turn up brightness in dark mode, the increase in power consumption is therefore inevitable. On OLED though, the power consumption is going to be related to how much of the screen is illuminated, and even this effect of turning up the brightness is unlikely to counteract that. Now it's possible I'm wrong and the effect does counterbalance (LED is less efficient at higher brightness for example), but what I find sloppy is that this could easily have been tested with an external OLED monitor, and while the paper itself is at least clear about that, the blog post just outright says dark mode uses more power, which is wrong, especially on phones where OLED is relatively more common than it is for monitors and dark mode is more popular.

...but what I find sloppy is that this could easily have been tested with an external OLED monitor, and while the paper itself is at least clear about that, the blog post just outright says dark mode uses more power, which is wrong, especially on phones where OLED is relatively more common than it is for monitors and dark mode is more popular.

It's the Bbc, where style is often greater than substance. What might save more energy is having their website stick to simple text & graphics. Some of their supposedly 'in depth' features have had an annoying habit of throwing in animations and other crap that requires scroll, scroll, scroll and waiting for it to drop in text/graphics instead of displaying a simple graphic.

Old/ordinary LCD displays have a single backlight illuminating the entire display. Black pixels turn the light into heat so displaying a full black screen uses as much power as full white one. Reasonably modern / older expensive displays have a grid of backlights so each light in the grid only needs to be as bright as the brightest pixel in front of it. Changing the background to black and using only a small window in one corner may save a little power at the display. With the display producing a tiny by less heat either your central heating has to work a tiny bit harder or your air conditioning does not have to work quite as hard.

I find bright text on a dark background much easier to read.

How "Responsiveness" affects power consumption depends entirely on how it's done.

If you go the easy way of just polling more often, then far more energy will be used.

To reduce energy consumption you have to be truly event driven and allow CPU cores to actually go to sleep, and not just sit there spinning in a tight loop doing nothing as fast as possible.

This is something I've been messing around with on my own PC. It's not a precise science, and day/night mode made virtually no difference for me. Obvious stuff like dropping display resolution did, but the slightly more suprising result was disabling a lot of the cruft that MS insists needs to be running in background. Killing as much of that as I dare reduced power consumption by 50-75W.

Which I guess is something I'm going to have to repeat now MS is insisting I 'update' to 24H2. Much has been written about power demands from 'AI' datacentres, but less from the combined impact of forcing 'AI' features like Copilot onto millions of PCs.

(Also curious if there are any power monitoring apps around that will show per-process consumption. Would be a neat feature for Task Mangler. That already has a cryptic green leaf status icon for processes that are in 'efficiency' mode, but MS isn't great at explaining exactly what that means.)

In all the years of using Dark Mode not once have I touched the brightness setting on my monitor !

Why would you?

The brightness is already set to what I want I use dark mode to stop the whiteout.

Guess I'm in the minority but yea ok.