let's contrast the sensibly designed device with the design i think you are describing
on the sensibly designed device, say i'm answering an email. i look at the screen, ponder for 30 seconds, scroll down, ponder for 30 seconds, and hit reply. redrawing the screen took, perhaps, a million clock cycles each time, so it required powering up a 240 megahertz esp32 for 2 milliseconds. so i've consumed 6 milliseconds of cpu, 0.6 millijoules at 100 milliwatts, plus the 50 microwatts of the screen for 60 seconds, which is 3 millijoules
then i spend 10 minutes typing out a 300-word reply at 60 words per minute with a duty cycle of about 50%. each time i press a key, the cpu powers up, processes for an average of 65536 clock cycles (0.3 ms), updates the screen, and powers back down. in total my 1500 keystrokes result in 450 ms of cpu time and 45 millijoules. keeping the screen active has cost another 30 millijoules.
finally i power up the 1.5 watt wifi radio hardware for 100 milliseconds and send the mail, consuming another 150 millijoules
total power consumption to read and answer the email: a bit under 250 millijoules. i'll get over 9000 emails out of a cr2032 coin cell, about six months of use
[i'm being very generous with these numbers to be sure to estimate conservatively. my best estimate is that this screen will hold 12 lines of 40 characters of text readably, and proportional-font layout with word wrap takes about 40 clock cycles per character, including blitting from a font bitmap, so redrawing the screen should take about 19200 cycles, not a million; and handling a text editing keystroke should average 7500 cycles, not 65536; and sending an email over wifi should require it to be powered up for 1 ms, not 100.]
with what i think is your proposed design, i have to leave the pi zero running at 1 watt for the whole 11 minutes, so i use 660 joules, 2640 times as much. the cr2032 coin cell will last me for 3.3 emails
2640 times more energy consumption means 1/2640 the battery life. with the battery they propose to use i think it's 33 emails vs. 90000 instead of 3.3 vs. 9000
[corrected obvious dumb math error above, sorry]
evidently the second device is not an adequate replacement for the first
even if it does support sixel
---
as for corrupting the flash, a flash block that was partially written before the chip lost power looks exactly like a correctly written block some percentage of the time, maybe a high percentage like 99%. but sometimes you will read back corrupted data, maybe depending on temperature, emi, or state of charge. so cutting power to your industrial sds when you're not sure if the flash is writing or not is going to bite you sooner or later
> evidently the second device is not an adequate replacement for the first
Agreed, but the 2nd one is available on preorder. I'd like to see the 1st one you describe, and I think it may happen. Buying the 2nd one to show my interest in the project seems like the best way to make that happen!
> even if it does support sixel
Well TBH if only the 2nd one supported sixels, that'd be my choice (though I'd work hard to have that working in the 1st one too lol)
> so cutting power to your industrial sds when you're not sure if the flash is writing or not is going to bite you sooner or later
When I type sync and wait a while, I do assume the flash is not writing.
on the sensibly designed device, say i'm answering an email. i look at the screen, ponder for 30 seconds, scroll down, ponder for 30 seconds, and hit reply. redrawing the screen took, perhaps, a million clock cycles each time, so it required powering up a 240 megahertz esp32 for 2 milliseconds. so i've consumed 6 milliseconds of cpu, 0.6 millijoules at 100 milliwatts, plus the 50 microwatts of the screen for 60 seconds, which is 3 millijoules
then i spend 10 minutes typing out a 300-word reply at 60 words per minute with a duty cycle of about 50%. each time i press a key, the cpu powers up, processes for an average of 65536 clock cycles (0.3 ms), updates the screen, and powers back down. in total my 1500 keystrokes result in 450 ms of cpu time and 45 millijoules. keeping the screen active has cost another 30 millijoules.
finally i power up the 1.5 watt wifi radio hardware for 100 milliseconds and send the mail, consuming another 150 millijoules
total power consumption to read and answer the email: a bit under 250 millijoules. i'll get over 9000 emails out of a cr2032 coin cell, about six months of use
[i'm being very generous with these numbers to be sure to estimate conservatively. my best estimate is that this screen will hold 12 lines of 40 characters of text readably, and proportional-font layout with word wrap takes about 40 clock cycles per character, including blitting from a font bitmap, so redrawing the screen should take about 19200 cycles, not a million; and handling a text editing keystroke should average 7500 cycles, not 65536; and sending an email over wifi should require it to be powered up for 1 ms, not 100.]
with what i think is your proposed design, i have to leave the pi zero running at 1 watt for the whole 11 minutes, so i use 660 joules, 2640 times as much. the cr2032 coin cell will last me for 3.3 emails
2640 times more energy consumption means 1/2640 the battery life. with the battery they propose to use i think it's 33 emails vs. 90000 instead of 3.3 vs. 9000
[corrected obvious dumb math error above, sorry]
evidently the second device is not an adequate replacement for the first
even if it does support sixel
---
as for corrupting the flash, a flash block that was partially written before the chip lost power looks exactly like a correctly written block some percentage of the time, maybe a high percentage like 99%. but sometimes you will read back corrupted data, maybe depending on temperature, emi, or state of charge. so cutting power to your industrial sds when you're not sure if the flash is writing or not is going to bite you sooner or later
better hope it's not on orbit when it happens
it was for me