The “Energy Impact” tab in OS X 10.9 Mavericks’ Activity Monitor is exactly what I would’ve wanted when I was testing my new Macbook Air a few months ago. Although, one thing I’ve found is that the “Avg Energy Impact” (it averages over the past 8 hours) isn’t actually a very useful/stable number, at least for measuring something that can have a very variable/spiky usage, like browsing with lots of tabs.
I knew if it was in Activity Monitor though, that the raw data must be available, and after some fruitless online searching, I just dug around my system and ended up finding some cool stuff in /usr/bin. First is a script called power_usage.sh that dumps out a report of lots of technical details.
Here’s a small sample of the output:
**** Battery and backlight usage **** Backlight level: 862 (range 0-1024) **** Network activity **** out: 1.00 packets/s, 94.29 bytes/s in: 1.50 packets/s, 212.53 bytes/s **** Disk activity **** read: 0.00 ops/s 0.00 KBytes/s write: 5.49 ops/s 32.70 KBytes/s **** Interrupt distribution **** CPU 0: Vector 0x46(SMC): 2.99 interrupts/sec Vector 0x49(MacBookAir6,1): 0.50 interrupts/sec Vector 0x92(IGPU): 107.76 interrupts/sec Vector 0x98(ARPT): 9.48 interrupts/sec Vector 0x9e(SSD0): 3.99 interrupts/sec Vector 0xdd(TMR): 691.96 interrupts/sec Vector 0xde(IPI): 8.98 interrupts/sec CPU 1: Vector 0xdd(TMR): 6.49 interrupts/sec Vector 0xde(IPI): 3.99 interrupts/sec CPU 2: Vector 0xdd(TMR): 16.46 interrupts/sec Vector 0xde(IPI): 8.98 interrupts/sec CPU 3: Vector 0xdd(TMR): 8.48 interrupts/sec Vector 0xde(IPI): 2.49 interrupts/sec **** Processor usage **** Intel energy model derived package power (CPUs+GT+SA): 0.58W LLC flushed residency: 93.2% System Average frequency as fraction of nominal: 85.75% (1457.75 Mhz) Package 0 C-state residency: 93.81% (C2: 13.45% C3: 0.07% C6: 0.00% C7: 80.29% C8: 0.00% C9: 0.00% C10: 0.00% ) Core 0 C-state residency: 96.16% (C3: 0.00% C6: 0.00% C7: 96.16% ) CPU 0 duty cycles/s: active/idle [< 16 us: 20.45/11.47] [< 32 us: 7.98/15.96] [< 64 us: 70.34/13.47] [< 128 us: 23.45/11.97] [< 256 us: 14.97/6.49] [< 512 us: 18.46/2.00] [< 1024 us: 4.99/4.49] [< 2048 us: 3.49/8.98] [< 4096 us: 2.00/12.97] [< 8192 us: 0.00/24.45] [< 16384 us: 0.00/26.94] [< 32768 us: 0.00/26.94] CPU Average frequency as fraction of nominal: 77.55% (1318.39 Mhz) CPU 1 duty cycles/s: active/idle [< 16 us: 15.47/4.49] [< 32 us: 1.00/2.49] [< 64 us: 2.99/2.00] [< 128 us: 1.50/0.50] [< 256 us: 4.49/1.00] [< 512 us: 1.00/1.50] [< 1024 us: 1.00/1.50] [< 2048 us: 0.50/0.50] [< 4096 us: 0.00/0.50] [< 8192 us: 0.00/2.49] [< 16384 us: 0.00/2.99] [< 32768 us: 0.00/0.00] CPU Average frequency as fraction of nominal: 83.51% (1419.63 Mhz) Core 1 C-state residency: 96.81% (C3: 0.01% C6: 0.00% C7: 96.81% ) CPU 2 duty cycles/s: active/idle [< 16 us: 68.35/4.49] [< 32 us: 8.98/5.49] [< 64 us: 5.49/17.96] [< 128 us: 6.49/12.97] [< 256 us: 7.98/13.47] [< 512 us: 2.99/7.48] [< 1024 us: 2.00/2.49] [< 2048 us: 1.00/2.49] [< 4096 us: 0.00/0.50] [< 8192 us: 0.50/7.98] [< 16384 us: 0.00/9.98] [< 32768 us: 0.50/7.98] CPU Average frequency as fraction of nominal: 94.05% (1598.86 Mhz) CPU 3 duty cycles/s: active/idle [< 16 us: 9.48/1.50] [< 32 us: 2.00/2.49] [< 64 us: 0.50/2.00] [< 128 us: 3.99/0.00] [< 256 us: 1.00/1.50] [< 512 us: 1.50/0.50] [< 1024 us: 1.00/0.50] [< 2048 us: 0.00/0.00] [< 4096 us: 0.50/0.50] [< 8192 us: 0.50/0.00] [< 16384 us: 0.00/2.99] [< 32768 us: 0.00/1.50] CPU Average frequency as fraction of nominal: 96.73% (1644.44 Mhz) **** GPU usage **** GPU 0 name IntelIG GPU 0 C-state residency: 99.73% (0.00%, 99.73%) GPU 0 P-state residency: 1100MHz: 0.00%, 1050MHz: 0.00%, 1000MHz: 0.00%, 950MHz: 0.00%, 900MHz: 0.00%, 850MHz: 0.00%, 800MHz: 0.00%, 750MHz: 0.00%, 700MHz: 0.00%, 650MHz: 0.00%, 600MHz: 0.00%, 550MHz: 0.00%, 500MHz: 0.00%, 450MHz: 0.00%, 400MHz: 0.00%, 350MHz: 0.28%, 300MHz: 0.00%, 250MHz: 0.00%, 200MHz: 0.00% GPU 0 average frequency as fraction of nominal (200.00Mhz): 0.49% (0.98Mhz) GPU 0 GPU Busy 0.28% GPU 0 FB Test Case 0.00%
Most of the best stuff is coming from /usr/bin/powermetrics. Here's the man page for powermetrics. There's a lot of options to go through.
I couldn't get some of the --hide flags to work, but here's a quick command to pull just the power score from some apps. Suffice to say, the possibilities for this in doing power comparison are quite tantalizing:
powermetrics -i 1000 --poweravg 1 | grep 'Average cumulatively decayed power score' -A 20
Example output:
--
**** Average cumulatively decayed power score ****
15 sec 1 min 5 min 15 min 1 hr
[26367]systemstats 157.144 43.3773 15.5405
[0 ]kernel_task 74.2694 74.8748 47.8391
[-1 ]DEAD_TASKS 48.8577 24.5274 6.72759
[123 ]WindowServer 30.0174 35.1636 21.6378
[388 ]iTerm 21.9219 20.0339 12.7875
[26917]Airmail 18.7339 19.6338 9.45401
[41829]Google Chrome He 10.6604 7.92617 4.61903
[42323]com.apple.WebKit 10.1828 17.2152 13.345
[41206]Safari 9.1336 16.8081 10.7422
[42352]Activity Monitor 7.60748 6.4298 3.80237
[96 ]hidd 7.59 7.70318 4.438
[42612]powermetrics 4.74147 4.81903 3.23933
[37370]SystemUIServer 4.36079 4.44251 3.02667
[159 ]sysmond 3.80035 2.93568 1.85484
[442 ]Moom 3.53523 2.59021 1.21046
[389 ]Dock 3.43152 2.84213 1.58266
I also made another cool discovery for power measurement. Originally I first tried to see if Apple's top had the power information, but sadly, it appeared like it didn't. Well, it turns out that (and the top man page) was wrong - my 11" MBA's screen simply wasn't big enough (even maximized) to show the POWER column, opening top open on my bigger desktop screen showed that this wasn't the case!
So here's an equivalent "power" command for top:
top -stats pid,command,power -o power -l 0
And the output looks something like:
PID COMMAND POWER 123 WindowServer 31.1 388 iTerm 25.7 42689 top 10.5 96 hidd 2.9 26917 Airmail 1.4 37370 SystemUIServer 0.4 45 fseventsd 0.3 85 mds 0.1 42323 com.apple.WebKit 0.1 31927- Dropbox 0.1 31997 WebKitPluginHost 0.1 453- Google Drive 0.1
It'd be pretty cool to make an open source power measuring/graphing framework combined w/ a browser automation test suite. I might get to it one day, but maybe someone w/ some more free time might feel like doing it sooner? Just putting it out there.