N900 Hardware Power Consumption

(Some preliminary numbers using the battery monitor chip.: added to note)
(Some preliminary numbers using the battery monitor chip.)
Line 31: Line 31:
* More media player measurements (ssh over wifi, gsm off, same file)
* More media player measurements (ssh over wifi, gsm off, same file)
  those values are moot - for the first 2 it's irrelevant which process is *idle* since the CPU should zeroclock anyway, alas here it's at fixed clock-freq which is NOT recommended. For the rest the only useful audio source to run those tests is a pink noise at 0dB. See `man sox` for how to create such file. Or to set volume to 1 step above zero, see above - so the whole music content becomes irrelevant. For arbitrary music the amplifier will introduce random "noise" up to 2000mW to the figures. The BQ24150 can't average out that noise. Further there's "GUI on" which means "backlight" and that's very random and even depending on ambient light, see 3 bullets above --[[User:joerg_rw|joerg_rw]] 12:52, 13 May 2015 (UTC)
  those values are moot - for the first 2 it's irrelevant which process is *idle* since the CPU should zeroclock anyway, alas here it's at fixed clock-freq which is NOT recommended. For the rest the only useful audio source to run those tests is a pink noise at 0dB. See `man sox` for how to create such file. Or to set volume to 1 step above zero, see above - so the whole music content becomes irrelevant. For arbitrary music the amplifier will introduce random "noise" up to 2000mW to the figures. The BQ24150 can't average out that noise. Further there's "GUI on" which means "backlight" and that's very random and even depending on ambient light, see 3 bullets above --[[User:joerg_rw|joerg_rw]] 12:52, 13 May 2015 (UTC)
 +
Not fixed, it's ondemand w/ bottom freqs allowed as specified.
 +
I restarted the same track every time, so the comparison should be quite  adequate, tho a noise would be better, yep.
 +
No backlight ofc.
 +
[[User:l29ah|l29ah]] 22:53, 14 May 2015 (UTC)
** idle @125MHz: 45mW T
** idle @125MHz: 45mW T
** idle @250MHz: 76mW T
** idle @250MHz: 76mW T

Revision as of 22:53, 14 May 2015

This page contains information on the amount of energy used by each subsystem of the N900, and where applicable ways it can be reduced.

Much of any reduction will require additional software, some may be possible by altering user behavior in ways that do not impact the users overall experience..

Measuring power usage is hard. For example, power consumption of the speakers/speaker-amp has to be disentangled from power consumption of reading a file from SD/eMMC, decoding it, PulseAudio's power use, and the DAC power use.


Below currents are from l3/l4 manual - assumed to be from a supply at 5 V.

  • FMTX 19 mA - l3/4
  • WLAN on - no powersave - 200 mA
  • WLAN transmit 802.11b - 150 mA

Battery parameters can be monitored thru fuel gauge chip TI BQ27200 [1]

Contents

Some preliminary numbers using the battery monitor chip.

These numbers are biased, and represent best cases. Some have had outliers which may be normal and unavoidable current spikes removed. This is under the assumption that the minimal power represents the minimum power achievable, and it should long-term be possible to - possibly with software patches - achieve these. +x indicates the measurement is for that element only. A raw measurement has no + (marked "T"), and is a total consumption of whole device, when operated in that state. Unless otherwise mentioned, numbers are for the idle locked case, with screen off, battery voltage "full" = 4.0 to 4.1V. Assume

  • Suspend (echo mem >/sys/power/state), wifi and GSM off. 2 mA T
  • Suspend, GSM on, connected to 3G (2 bars) 5 mA T
  • Suspend, GSM on, connected to 2G. (2 bars) 3.6 mA T
  • Idle, no SIM, wlan off 7 mA@4.1 V T
  • Wifi on with maximum powersaving. +~2 mA
  • BT enabled, no headset connected +2 mA with ~1/5 time spikes to +9 mA
  • MMC card in. Under +1 mA
  • Xchat running, sporadic traffic. +8 mA
  • Screen on, no backlight. +80 mA
  • Backlight on max +150 mA, +40mA on 4 of 5 steps in simple brightness applet & low ambient (/sys/class/backlight/acx565akm/brightness = 63)
  • Media player playing MP3, one step above mute. 110 mA T
  • More media player measurements (ssh over wifi, gsm off, same file)
those values are moot - for the first 2 it's irrelevant which process is *idle* since the CPU should zeroclock anyway, alas here it's at fixed clock-freq which is NOT recommended. For the rest the only useful audio source to run those tests is a pink noise at 0dB. See `man sox` for how to create such file. Or to set volume to 1 step above zero, see above - so the whole music content becomes irrelevant. For arbitrary music the amplifier will introduce random "noise" up to 2000mW to the figures. The BQ24150 can't average out that noise. Further there's "GUI on" which means "backlight" and that's very random and even depending on ambient light, see 3 bullets above --joerg_rw 12:52, 13 May 2015 (UTC)
Not fixed, it's ondemand w/ bottom freqs allowed as specified.
I restarted the same track every time, so the comparison should be quite  adequate, tho a noise would be better, yep.
No backlight ofc.
l29ah 22:53, 14 May 2015 (UTC)
    • idle @125MHz: 45mW T
    • idle @250MHz: 76mW T
    • mediaplayer mp3 playback, max volume, speakers: 415mW T
    • mediaplayer mp3 playback, max volume, headphones: 262mW T
    • openmediaplayer mp3 playback, max volume, headphones: 230mW T
    • mediabox mp3 playback, max volume, speakers: 463mW T
    • mediabox mp3 playback, max volume, headphones: 324mW T
    • oscp mp3 playback, 99% volume, gui on, speakers: 434mW T
    • oscp mp3 playback, 99% volume, gui on, headphones: 301mW T
    • oscp mp3 playback, 99% volume, gui off, headphones: 273mW T
    • oscp mp3 playback, ALSA, PA's mixer setup, 99% softvol, gui off, headphones: 195mW T
  • Vibrator at max +120 mA
  • md5sum /dev/zero 250 mA T
  • md5sum /dev/mmcblk1p1 - 1G SD card. 240 mA T
  • md5sum /dev/mmcblk0 - internal MMC 240 mA T (these indicate the CPU is somewhat idle, waiting on IO, and this counteracts any increased power use by the memory)
  • camera active, showing preview. (backlight off) 210 mA T
  • GPS - location test gui - in continuous GPS recieve mode - +50-80 mA
  • 3G connected, good signal, no data or other activity +8 mA.
  • 2G connected, good signal, no data or other activity +5 mA

2x bash infinite loops, by frequency

  • 250MHz 278mW T
  • 500MHz 524mW T
  • 805MHz 903mW T
  • 900MHz 1064mW T

So it seems like the power saving cpufreq hacks are largely irrelevant and one would always use ondemand for maximum CPU peak throughput. Needs further testing with irregular I/O-dependent load tho.

NOTE: everything >500MHz WILL destroy your CPU long term, due to overheating and EM This is as evident as the rule "You shall not operate 12V incandescent Lamp at 15V or 20V" --joerg_rw 12:28, 13 May 2015 (UTC)

2G data, short pings

(all Total - ATTENTION, these values depend on distance from device to base station and thus how much TX RF power it needs)

  • 1 s ping 130 mA
  • 10 s ping 70 mA
  • 30 s ping 45 mA

3.5G data, short pings

  • 1 s 186 mA
  • 10 s 120 mA approx.
  • 30 s 136 mA

3.5g data, long pings (8000 bytes)

  • 1 s 240 mA
  • 10 s 205 mA
  • 30 s 120 mA

2G data, long pings (8000 bytes) (with specified seconds between pings)

  • 1 s 160 mA
  • 10 s 105 mA
  • 30 s 50 mA

3G sending file at 150 k/s 375 mA 3G recieving at 200 k/s 275 mA

random adds:

WLAN PS-mode medium,10 mW, ssh login active, local xterm with bq27200 script loop -9, screen blanked, aggressive custom breathing indicator light: 20 mA

xchat 8 freenode chan, wifi PSM medium, screen locked:

Nokia-N900-42-11:~# bin/bq27200.sh 20
LOOPMODE=20
      mv   RSOC CSOC mA   NAC  CACD CACT TTF   TTE   TEMP EDV1 LOW
18:36 4092 100  100  -134 1187 1187 1187 65535 530   37  0
18:36 4092 100  100  -17  1186 1186 1186 65535 3990  37  0
18:37 4106 99   99   -68  1186 1186 1186 65535 1042  37  0
18:37 4095 99   99   -52  1185 1185 1185 65535 1359  36  0
18:37 4130 99   99   -18  1185 1185 1185 65535 3807  36  0
18:38 4092 99   99   -138 1184 1184 1184 65535 511   36  0
18:38 4090 99   99   -110 1184 1184 1184 65535 644   36  0
18:38 4090 99   99   -57  1183 1183 1183 65535 1241  36  0
18:39 4090 99   99   -118 1183 1183 1183 65535 597   36  0
18:39 4114 99   99   -34  1182 1182 1182 65535 2077  36  0
18:39 4127 99   99   -17  1182 1182 1182 65535 3569  36  0
18:40 4122 99   99   -77  1181 1181 1181 65535 570   36  0
18:40 4087 99   99   -64  1180 1180 1180 65535 3403  36  0

A script to monitor the power consumption for a minute using bq27x00_battery

The shell script using i2cdump affects measurements a lot so it's better to use the kernel module.

CITATION NEEDED! Why should a shell script and i2cget be inferior to a perl script and a deprecated kernel module? --joerg_rw 12:18, 13 May 2015 (UTC)

apt-get install perl
#!/usr/bin/perl
use strict;
use warnings;
use List::Util qw(sum);
sub mean { return @_ ? sum(@_) / @_ : 0 }
my @pa;
for (my $i = 0; $i < 15; $i++) {
	open(my $cf, '<:encoding(UTF-8)', '/sys/class/power_supply/bq27200-0/current_now');
	open(my $vf, '<:encoding(UTF-8)', '/sys/class/power_supply/bq27200-0/voltage_now');
	my $cc = <$cf>;
	my $vc = <$vf>;
	my $pc = $cc * $vc / 1000000000; # mW
	push @pa, $pc;
	sleep(4);
}
print mean(@pa);

Software

Many software subsystems interact to aim to reduce battery usage.