Kleidos
EN ES
Follow the project
Menu

Measurements

LilyGO T-Deck Plus

With Kleidos on the LilyGO T-Deck Plus and the 2,000 mAh cell I measured it with, my estimate is that a charge lasts about 11 to 21 days. Locked and asleep it draws 3.93 mA; on the vault list at the 85 percent brightness it ships at, 85.4 mA.

Every current on this page comes from one LilyGO T-Deck Plus with the measurement system in its battery wire, on a development build that runs the logging, the console and the profiler. Each comes from a single run except the sleep figure, the median of three runs that agree with one another. The times further down I took on the same board with its own clock, on USB power, at three vault sizes from a small one to a full vault. I have not yet measured what typing to a connected computer draws, or what the cell really holds.

Kleidos on the LilyGO T-Deck Plus: its screen, what each control does, what it holds and what it costs

Updated

How long a charge lasts: an estimate

My estimate of the days one charge lasts. I worked it out from the states in the next section; I have not measured a charge running down.

Each session is 15 seconds on the PIN screen, 30 on the vault list and 10 on the Bluetooth typing screen, then the idle timers: 10 seconds lit and 50 dark before the board goes back to sleep, 115 seconds awake in all. I work the days out for four ways of using it, from 0 to 50 sessions a day, with the board asleep for the rest of every day.

Use Sessions a day Days per charge (estimate)
Left in a drawer 0 21
A few times a day 5 19
About once an hour, all day 20 15
Heavy use 50 11

It is the sleep that empties the cell, not the use: asleep it uses 94.2 mAh a day, and by my estimate a session adds 1.9 mAh.

What goes into those days, and which way each input moves them:

State by state, against the budgets

What the board draws in each state it passes through, from asleep with the vault locked to the admin network up, every figure taken with the measurement system. The bar under each figure is on one scale for the whole page. Where the project set itself a budget, it stands beside the figure, and a figure over it says so in words.

Each margin is the measurement system's alone, for the one run behind its figure, with the shunt's tolerance assumed. The store holds repeat runs of some of these states, and I have worked out their spread only for the sleep figure, whose card prints no margin, so no margin here says where a second run would land.

Measurement system: an INA228 in the battery wire, read by an ESP32-S3, with the USB cable left plugged in but its port switched off at the hub, so the cell alone powers the board.

What the screen's brightness costs

The vault list at each brightness level, so the setting can be weighed in milliamps. Each level is a measuring window of its own, taken in a different run from the vault list's figure above and at a slightly lower cell voltage, which is why the two readings at 85 percent differ a little.

Chart of measurement system readings

What the board draws on the vault list at each brightness level I measured: one dot per level, on a milliamp scale that starts at zero. The firmware's setting starts at 10 percent, and I measured the levels below it too. The line only joins the dots, because I measured nothing between two levels, and every value is in the table folded under the chart.
See the 21 values
Brightness Board draws (mA)
0 % 51.5
5 % 53.6
10 % 55.6
15 % 57.7
20 % 59.7
25 % 61.6
30 % 63.6
35 % 65.6
40 % 67.7
45 % 69.5
50 % 71.7
55 % 73.7
60 % 75.7
65 % 77.6
70 % 79.7
75 % 81.7
80 % 83.7
85 % 85.6
90 % 87.6
95 % 89.6
100 % 91.7

From 0 to 100 percent brightness, the board draws 51.5 to 91.7 mA, and 85.6 mA at the 85 percent it ships at.

At 0 percent the light is off but the panel is still awake; with the panel asleep and the chip kept out of light sleep, the board draws 45.8 mA.

What the keyboard light costs

What lighting the keyboard adds to the board's draw at each of its levels. I measured every level on the PIN screen against the light off just before and just after it, in the same run, so no figure is a difference between two runs. Each is the mean of 2.3 seconds with the light on, and it includes the little extra the rest of the board draws while the light pulls the battery's voltage down.

Chart of measurement system readings

What the keyboard light adds to the board's draw at each level I measured: one dot per level, on a milliamp scale that starts at zero. The light ships off, so no level is ringed as the one it ships at. The line only joins the dots, because I measured nothing between two levels, and every value is in the table folded under the chart. The curve is one run; its margin at 100 percent is ±1.6 mA, with the shunt's tolerance assumed, and I have worked out no margin for the other levels.
See the 20 values
Light level Added by the light (mA)
5 % 5.9
10 % 12.0
15 % 18.4
20 % 24.8
25 % 30.7
30 % 37.2
35 % 43.8
40 % 50.3
45 % 56.5
50 % 63.2
55 % 70.0
60 % 76.7
65 % 83.2
70 % 89.9
75 % 96.9
80 % 104.1
85 % 110.4
90 % 117.5
95 % 124.8
100 % 132.5

From 5 to 100 percent, the keyboard light adds between 5.9 and 132.5 mA to what the board draws with it off. Turned on, it lights at 50 percent for 8 seconds after each keypress until another level is chosen, and adds 63.2 mA while it is lit.

How long an unlock and a backup take

How long the board keeps a person waiting. I timed it with the board's own clock, not with the measurement system. Each column is the vault it held, in credentials and 2FA codes: the small vault that is the same on every board, then half of what this board holds, then all of it.

What was timed 30 / 10 (ms) 250 / 25 (ms) 500 / 50, full (ms)
Unlock From the passphrase entered to the vault list on screen 2,867 3,446 3,496
Of that, opening the vault From the derived key to the vault open 691 1,186 1,373
Wrong passphrase From the passphrase entered to its refusal 1,617 1,702 1,564
Encrypted backup to microSD From its key's derivation to the last byte written 6,201 13,464 16,413

The key derivation alone, at 175,000 iterations, takes 1,423 ms whatever the vault holds, since it opens none.

Each figure is the mean of its repeats on the board's own clock, rounded to the millisecond: ten unlocks, four wrong passphrases and three backups at each size, and five derivations. The unlock, the vault's opening and the refusal are each the difference between two of the board's log lines, stamped to the millisecond; the firmware times the backup itself, in whole milliseconds; and the derivation alone is read off the chip's microsecond counter. The board's crystal is not marked, so I take its tolerance as ±10 ppm, the figure the chip maker's design guidelines give.

What the figures include and leave out:

What I have not measured on it yet

Each of these is a figure this page would print and does not, because I have not taken it yet.

Where these figures come from

The first list names files in the repository, which I keep private while I prepare the first release; the second links pages you can open now.

The records behind the figures on this page are published as data under CC BY 4.0, byte for byte as the measurement store holds them. The CSV gives every value in them, with the address and the SHA-256 of its record, and each record that a card above names is linked from its name.