Power consumption¶
First real measurement of what the Tab5 draws, made on 2026-10-10 on a tablet with its battery by husyildiz (firmware 3.8.0-rc.4, discussion #278), with the Home Assistant script « Tab5 — consumption test » (packages/tab5_mesure_conso.yaml). Thanks to them for the hour of patience.
How it was measured¶
- Eleven cases of 5 minutes each (the first 45 s of each are not counted), the tablet unplugged and left alone, 47 to 51 readings per case.
- The power is the battery voltage × the battery current read by the INA226, so it is the power at the battery, not at the wall. A charger has losses of its own, not measured here.
- One tablet (display chip ST7121), one run, a battery between 92 and 100 %. The reference case, played again at the end, gave the same value (2.63 W): no drift. Inside a case the readings vary by about ±0.05 W: differences smaller than that mean nothing.
Results¶
| Case | Average | vs reference | Full battery would last (14.8 Wh rating) |
|---|---|---|---|
| Reference: screen 100 %, Okay Nabu on, speaker on | 2.63 W | 5.6 h | |
| Okay Nabu (microphone) off | 2.58 W | −0.05 W | 5.7 h |
| Speaker amplifier off | 2.58 W | −0.05 W | 5.7 h |
| Screen 50 % | 1.74 W | −0.89 W (−34 %) | 8.5 h |
| Screen 10 % | 1.73 W | −0.90 W | 8.6 h |
| Energy saving « Always » | 1.73 W | −0.90 W | 8.6 h |
| Screen off | 1.23 W | −1.40 W (−53 %) | 12.0 h |
| Dark theme (Obsidienne), screen 100 % | 2.63 W | 0.00 W | 5.6 h |
| Same theme, light mode | 2.65 W | +0.02 W | 5.6 h |
| Screen, Okay Nabu and speaker all off | 1.16 W | −1.47 W (−56 %) | 12.8 h |
| Reference again (end of the run) | 2.63 W | 0.00 W | 5.6 h |
The battery's rating is 7.4 V × 2000 mAh = 14.8 Wh (M5Stack documentation, via its distributors); M5Stack itself gives about 6 hours at 50 % brightness with Wi-Fi and background tasks running. The hours above are a rating divided by a power: an upper bound, not a measured autonomy. The script printed 23.7 Wh as « rough usable capacity »: that figure comes from only 7 points of battery level (a level computed from the voltage, which is not linear for a lithium cell) and exceeds the rating, so it is not to be trusted. The hours it printed (9 h in the reference case) are too high.
What it tells¶
- The backlight is the only big lever. 100 % → 50 % saves 0.89 W, one third of the total. Switching the screen off saves 1.4 W, more than half.
- Below 50 % there is nothing more to gain: 50 % and 10 % give the same power (1.74 and 1.73 W). The cause is not established (a floor of the backlight driver is the first guess); a measurement at 75 % and 25 % would tell.
- The energy saving mode is, in practice, its 50 % cap. « Always » (50 % cap, 10 % after 30 s without a touch, no animation, 30 fps) draws the same as the screen at 10 %: no measurable gain from the animations or the frame rate.
- The theme changes nothing: dark or light, 0.02 W. The screen is backlit, so a black pixel does not switch the light off (the panel type was not checked against its datasheet). A dark theme on battery is for taste, not for autonomy.
- Voice costs almost nothing. The microphone (Okay Nabu) and the speaker amplifier each draw about 0.05 W, which is the noise of the measure. Switching Okay Nabu off to save power gives up the voice assistant for nothing.
- The rest of the tablet draws 1.2 W with everything off (processor, Wi-Fi through the ESP32-C6, memory): the floor of the whole device as it runs today.
Over a year¶
Power × 8.76 gives the kWh of a year of continuous running (at the battery, charger losses not included):
| State, all year | Power | kWh per year |
|---|---|---|
| Screen 100 %, voice on | 2.63 W | 23.0 |
| Screen 50 %, voice on | 1.74 W | 15.2 |
| Screen off, voice on | 1.23 W | 10.8 |
A day with the screen lit for 6 hours and off the other 18 (an assumption, to adapt to your use; the voice stays on and wakes the screen):
| Profile | Average | kWh per year |
|---|---|---|
| Screen 100 % all day | 2.63 W | 23.0 |
| 100 % for 6 h, then off | 1.58 W | 13.8 |
| 50 % for 6 h, then off (the « eco » profile) | 1.36 W | 11.9 |
The eco profile uses about half of the always-lit reference. In money, multiply the kWh by your price: at 0.25 €/kWh (an example, not a quote), 11.9 kWh is about 3 € a year, and the reference about 5.75 €. The savings are real but small in euros: the point of the profile is the heat, the battery wear and the ease of living with a wall screen that dims itself.
Waveshare 7B, at the USB input (2026-10-16)¶
The second board of the firmware (Waveshare 7B, 7 inches, 1024×600) measured by the author, the board powered by a computer's USB port.
How it was measured¶
- A FNIRSI FNB58 USB tester between the supply and the board, read by the computer through its micro-USB « PC » port: 100 samples per second (about 37 new values per second, the others repeat), voltage and current to 0.01 mV / 0.01 mA. The power is measured at the board's USB input, so it includes its regulators; the Tab5 figures above are at its battery. The tester's accuracy is not given by its maker.
- The board is driven by its own API while the tester records (
tools/campagne_mesures.py, readertools/fnb58.py): 20 s per backlight level after 6 s to settle; for each of the 20 screens of the « Aller à l'écran » select, the first 3 s (opening) then 15 s once open; three restarts with the « Redémarrage Système » button, 50 s each. No measure falls within −5 s / +30 s of a minute that is a multiple of 5, when Home Assistant sends its grouped updates. Everything is put back at the end. - Firmware
main46ad7f03, theme Ardoise douce in dark mode, full animations, Okay Nabu and the speaker on, Wi-Fi eco off, the home screen. One board, one run (156,988 samples). Inside a segment, the power varies by about ±0.08 W (median to 95th percentile): a difference of a few hundredths of a watt between two segments means nothing.
Results¶
| Case | Average | Current at ~5 V |
|---|---|---|
| Home screen, backlight 189/255 (74 %, the author's setting) | 1.96 W | 393 mA |
| Backlight 100 % | 3.14 W | 645 mA |
| Backlight 75 % | 1.98 W | 397 mA |
| Backlight 50 % | 1.47 W | 293 mA |
| Backlight 25 % | 1.28 W | 252 mA |
| Backlight 10 % | 1.25 W | 246 mA |
| Backlight 5 % | 1.24 W | 246 mA |
| Backlight 1 % | 1.24 W | 245 mA |
| Backlight off | 1.24 W | 245 mA |
| Each of the 20 screens and popups, open (at 74 %) | 1.945 to 1.958 W | 390 to 393 mA |
| Opening a screen (first 3 s) | 1.957 to 1.987 W, peak 2.28 W | |
| A restart (50 s from the button) | 1.93 W on average, 26.7 mWh |
A restart, second by second: the board stops about 2 s after the button, drops to about 1.1 W for 1.5 s (the chip resets, lowest sample 0.78 W), runs at about 1.8 W while it builds the screen, about 2.0 W once the Wi-Fi starts (around 7 s), and is back to 1.96 W after about 14 s; highest sample 2.57 W. Two restarts gave the same curve to ±0.02 W. The third one did not finish: the Wi-Fi chip did not come back, see the 7B page.
What it tells¶
- The last quarter of the backlight costs the most. From 75 % to 100 % the board draws 1.16 W more; from off to 75 %, only 0.74 W. The light given out was not measured (no light meter): whether 100 % looks much brighter than 75 % is for the eye to judge.
- Below 25 %, nothing more to gain: 25 % draws 0.04 W more than the backlight off. On the Tab5 the same floor started below 50 %.
- Drawing costs nothing measurable. Every screen and popup, open, draws the same as the home screen (within 0.013 W, the noise); opening one adds at most 0.03 W over 3 s. Choosing a lighter screen for power saving makes no sense.
- A restart costs nothing either: 1.93 W on average over the 50 s, less than a lit screen, because the backlight is off while the chip resets.
- The board's floor is 1.24 W with the backlight off (processor, Wi-Fi through the ESP32-C6, microphone and speaker on): the same as the Tab5 at its battery (1.23 W), by chance or not.
- The USB voltage sags at full brightness: 4.87 V at 645 mA through the tester and a computer port, against 5.06 V with the backlight off. Not a problem here, but a long or thin cable or a weak supply leaves less margin at 100 %.
Tab5 and 7B side by side¶
Not the same measuring point (battery for the Tab5, USB input for the 7B), not the same screen size (5 and 7 inches): an order of magnitude, not a contest.
| Tab5 (at the battery, 2026-10-10) | Waveshare 7B (at the USB input, 2026-10-16) | |
|---|---|---|
| Backlight 100 % | 2.63 W | 3.14 W |
| Backlight 50 % | 1.74 W | 1.47 W |
| Screen off | 1.23 W | 1.24 W |
| No gain below | 50 % | 25 % |
| First image after a restart | about 21 s (before #478; not measured since, the Tab5 is out of order) | 5.6-5.8 s (with #478) |
Over a year, 7B¶
| State, all year | Power | kWh per year |
|---|---|---|
| Backlight 74 % | 1.96 W | 17.2 |
| Backlight 50 % | 1.47 W | 12.9 |
| Backlight off | 1.24 W | 10.9 |
| 74 % for 6 h, then off | 1.42 W | 12.4 |
| 50 % for 6 h, then off | 1.30 W | 11.4 |
USB supply losses not included (they were not measured).
Settings and themes, 7B (night of 2026-10-16)¶
Same tester, same board, the night of the speed measurements (21:45 → 06:16, 3,065,400 samples, no gap): a measuring build of main 3293d4e1, backlight 189/255 (74 %), the home screen, 60 s per case (20 s per theme). The reference is Ardoise douce in dark mode, full animations, Okay Nabu and the speaker on, Eco Wi-Fi off.
| Case | Average | vs reference |
|---|---|---|
| Reference, start / end of the night | 1.956 / 1.957 W | |
| Each of the 21 themes, dark | 1.944 to 1.968 W | none |
| Each of the 21 themes, light | 1.995 to 2.013 W | about +0.05 W (+2.5 %) |
| Obsidienne, dark / light | 1.955 / 2.003 W | light +2.5 % |
| Animations « None » / « Essential » | 1.950 / 1.959 W | none |
| Power saving « Always » | 1.238 W | −37 % |
| Eco Wi-Fi « Always » | 1.781 W | −9 % |
| Okay Nabu off | 1.938 W | −1 % |
| Speaker off | 1.946 W | −0.5 % |
| Microphone and speaker off | 1.905 W | −2.6 % |
| A restart (10 restarts, 40 s each from the button) | 1.920 to 1.931 W | |
| The night, screen off (23:32 → 00:02) | 1.235 W | |
| The night, screen on (00:03 → 06:15) | 1.957 W |
- Neither the theme nor the animations change the power. The 21 dark themes are within 0.024 W of each other (the noise); light mode adds about 0.05 W, whatever the theme.
- Power saving « Always » brings the board to its floor: 1.238 W, the same as the backlight at 1 % or off. On a home screen left alone, this mode lowers the brightness to the minimum after 30 s without a touch (settings). Whether the screen stayed readable was not checked.
- Eco Wi-Fi « Always » saves 0.18 W (−9 %), about 1.5 kWh a year if left on all the time; answers may come a little later, which was not measured.
- The microphone and the speaker cost little: 0.05 W for both.
- No drift over the night: 12.79 Wh in 6 h 43 min (1.90 W on average, half an hour of it with the screen off); with the screen on, the hourly average stayed between 1.956 and 1.958 W.
What to do with it¶
Settings that already exist: Brightness (50 % is the sweet spot), Auto screen off (5 min), Power saving (« On battery » is the default), Charge limit 80 % for a tablet that stays plugged in, Wake the screen on « Okay Nabu » and with a tap to get it back.
Not worth doing, according to the measure: dark theme on battery (0 W), switching Okay Nabu off (0.05 W and no voice).
Ideas that would change something (not done, to be decided):
- On battery, an automatic screen off after 5 minutes, today independent of the power source: it takes 1.74 W down to 1.23 W while nobody looks at the tablet.
- Measure the missing points: screen at 75 % and 25 % (why does nothing change under 50 %?), the Eco Wi-Fi, and the wall power with a smart plug (charger losses, with and without a battery).
- Make the consumption test script compute the autonomy from the battery's rating instead of from its own extrapolation.
Against the usual solution (an old tablet)¶
This project has not measured another tablet: no figure is given here for one, and none found in a search was backed by a method. To compare, put a measuring plug (Shelly, Tapo P110, or the like) under the tablet for 24 hours with the screen lit, then compare with 1.7 to 2.6 W above (to compare fairly, add the Tab5's charger, also measured at the wall).
What can be said without a measure:
- The Tab5 runs a native program, not a browser: nothing to update on the tablet side except this firmware, no kiosk application, no Android version that stops being supported.
- Its battery is optional (it runs on USB alone, as the author's does), and the charge limit holds a plugged-in battery at 80 %. A swollen battery pushing out the screen of an old tablet left on its charger for years is a well-known complaint (Home Assistant forum, SmartThings forum).
- Push-only: the tablet never polls, so the Wi-Fi and the processor have little to do between two changes.
Adapting it to your home, with or without an AI¶
Everything that depends on a home is set in Home Assistant, not in the firmware: rooms and tiles, climate, energy, cameras, tracked sensors, the shortcuts of the gestures. The firmware names no entity and calls no Home Assistant action (ADR-0025): it sends events, and the packages map them to a whitelist. Seven screen languages, themes, popups, the settings above are all choices of the user. And the project is written to be changed by an AI: AGENTS.md tells any agent what to read first, each file opens with an [AI-CONTEXT] header, and the tests refuse a change that breaks the contract with Home Assistant. This page itself comes from a measure taken by a user and analysed with Claude the same day.