Pandora power consumption breakdown

[_wb_]

In case you are wondering what the power consumption of various components of a Pandora is: I did some measurements using my System Info application to figure that out.


The Pandora battery has a design capacity of 4200 mAh, or about 16 Wh at 4 volt. Because of battery wear, let's say most people will get about 15 Wh out of their battery, so if you're consuming 1W that means it lasts 15 hours.


I can't really measure directly how much the consumption is when suspended or switched off. I guess I could estimate it by checking how much charge got lost, but I haven't done that yet (I never suspend or switch off my Pandora).


With the lid closed and no cpu usage, but turned on (standby mode), the Pandora uses about 0.11 W. That's the base level, which cannot be avoided as long as it's running.


The display itself (not the backlight) takes about 0.39 W. The backlight takes about 0.58 W if it's at max brightness, 0.1 W at the lowest brightness (besides "off").


So in total, with an idle cpu, lid open, it takes about 0.11+0.39+0.1 = 0.6 W at lowest brightness, 1.08 W at highest brightness.


Wifi on takes about 0.05 W if it's just sitting idle (maintaining its connection), and about 1 W if it's sending or receiving data.


Sound doesn't seem to make a big difference (except of course the cpu it takes) - I wonder why.


The CPU power consumption depends on its clock speed and its usage: I only have a Rebirth model so I don't know how the 1GHz model performs. The CPU seems to be the most efficient (per MHz) at 500 MHz: it uses about 0.7 W if used 100% at 500 MHz. At a lower clock speed it uses less power, e.g. about 0.5 W if used 100% at 350 MHz. At a higher clock speed it uses more power, e.g. about 1 W at 650 MHz, 1.5 W at 850 MHz. Note of course that in typical usage, the CPU will also be used less if you clock higher; these numbers are for 100% CPU usage.


So here are some estimated total power consumptions for a number of situations:


- Standby: 0.11W or ~136 hours


- Word processing / coding / reading stuff: ~0.6 W or 25 hours at low brightness, ~1.08 W or ~14 hours at max brightness


- Light browsing (like reading mails / forums / blogs, no heavy downloading or cpu usage most of the time): ~0.8 W or ~18 hours at low brightness


- Light gaming (assuming 80% cpu at 500 MHz): ~1.2 W or 12.5 hours at low brightness, ~1.7 W or 8.8 hours at max brightness


- Heavy gaming (assuming 100% cpu at 850 MHz): ~2.1 W or 7.1 hours at low brightness, ~2.6 W or 5.8 hours at max brightness


- Heavy gaming while doing heavy downloading: ~3.1 W or 4.8 hours at low brightness, ~3.6 W or 4.2 hours at max brightness


Conclusion: the 10 hour battery life claim holds under reasonable assumptions, but of course not if you really use everything to the max.

 

[frostfall]

I've always wondered why my pandora last for like 6->7h and not the 10h annonced... It's because when my pandora is on, I play heavy games (like android ones) with wifi on (but I'm not downloading)


That's a kind of relief, because I've always thought since I received my pandora that I had a bad battery or whatever...It's just that "10 hours" is real only with light gaming and no wifi...

 

[_wb_]

Backlight brightness makes quite a big difference, and of course the CPU usage. If you're playing CPU-heavy games and want to get more out of your battery, you can try with less overclocking and a lower OPP setting - if you can live with a lower number of fps. Note that clocking your CPU to 850 MHz makes it use twice as much power as clocking it to 500 MHz. So if your game runs well enough at 500 MHz, try that. It may not use the CPU 100% of the time at the higher clockspeed, but it will probably still keep it awake in C0 most of the time (it's unlikely that the CPU can really ever go to the deep sleep idle states like C3 or C5).

 

[klapse]

Thanks for the informative post.

With the lid closed and no cpu usage, but turned on (standby mode), the Pandora uses about 0.11 W. That's the base level, which cannot be avoided as long as it's running.

While 0.11W sounds low at first, that is 110mW or 22 times more than my idle N900 consumes.


This is not an issue if you use your Pandora every other day, but for occasional users, one would like to be able to put the device on a shelf and have almost yero battery drain.


I wonder what the difference to the N900 is? Are some system components (like the DAC or amplifier) not getting powered-down?

 

[Exophase]

The battery consumes about 3.7V nominally, not 4V. 4V is just what you measure when it's fully charged. Li-Ion batteries follow a discharge curve that's close to flat for most of its discharge period, and close to 3.7V.


No idea how wear effects this.

 

[_wb_]

Thanks for the informative post.

With the lid closed and no cpu usage, but turned on (standby mode), the Pandora uses about 0.11 W. That's the base level, which cannot be avoided as long as it's running.

While 0.11W sounds low at first, that is 110mW or 22 times more than my idle N900 consumes.


This is not an issue if you use your Pandora every other day, but for occasional users, one would like to be able to put the device on a shelf and have almost yero battery drain.


I wonder what the difference to the N900 is? Are some system components (like the DAC or amplifier) not getting powered-down?

Well 0.11W could be a slight overestimate because that's with the system monitor still running, waking up the cpu every second to measure everything.


The DAC get powered down when not in use, so that's not it. The power on indicator led probably takes some power, but it should not be that much. Also the RAM takes some power and the cpu too. I'll try to estimate the power use when suspended, that should be a bit lower than lid closed standby.


@Exophase: my measurements were made by multiplying the reported current (in Amp) by the reported voltage (in Volt) to get the power consumption in Watt. The estimate of 15Wh for a normal full battery should be OK, at 3.7 V that is about 4050 Ah and the design capacity is 4200 Ah. I think you can charge brand new batteries even higher than their design capacity: I think I recently charged my spare battery (which had seen very little use) to about 4500 Ah. With a not so fresh battery you can probably still get 13-14 Wh.

 

[_wb_]

While suspended to ram, power consumption seems to be between 65 and 70 mW. That is not much lower than the lid closed idle cpu power consumption of about 110 mW.


As far as I understand things, the only difference between lid closed and suspend is that when suspended, the cpu is completely powered off, while with just the lid closed, it is only "sleeping" in some idle state. Also the "power on" led is dimmed, which saves some additional power.


So I guess that 70 mW is for keeping the ram powered, the dim "power on" led, and I don't know what else.


The "power on" led eats about 18 mW when it's fully on (which is still the case when the lid is closed!). If you really want to get the highest possible battery life, you may want to consider not using that led (and the other ones for that matter). In normal use, 18 mW is not significant, but in lid closed standby mode, it makes quite a big difference: ~109 mW with the led on, ~90 mW with the led off, that's the difference between 138 hours (5.7 days) of standby time and 167 hours (6.9 days). One day difference is a big difference. So: suggestion to the SZ maintainers: please change the lid-close script to dim or turn off the power led on lid close! Oh and while you're at it: make the lid close script underclock the cpu to 125MHz (best to check first if the cpu is used more than, say, 5%, because if it is, the user probably doesn't want the cpu to be underclocked because he's playing music or something). At 125MHz and with the power led off, I got my "lid closed" power consumption down to ~80 mW, which means ANOTHER day of extra standby time (188 hours, 7.8 days).


Actually it looks like there is not much additional power saved by suspending to ram. Maybe suspend to SD would be a good idea?

 

[Exophase]

Actually it looks like there is not much additional power saved by suspending to ram. Maybe suspend to SD would be a good idea?

As an option, perhaps. I get the feeling that it wouldn't be much faster than powering back on from scratch if you don't actually need to have the state saved any more than the programs you're running can save it themselves.


Actually that makes me wonder, is there a signal everyone gets when the OS is about to power down due to not having enough battery? I feel like I should already know this..

 

[_wb_]

It probably just kills everything as usual when it wants to power down, first with 15 (SIGTERM) and after some timeout with 9 (SIGKILL).


Why do you think suspend to SD would be that slow? Depending on the speed of the SD card, writing or reading 512MB (a memory dump) should take not more than 1 second, and isn't that essentially the only thing that needs to be done? Or am I being naive here?

 

[Exophase]

*squints*


You want to reread what you just wrote? You just claimed that you can achieve 512+MB/s write speeds on Pandora's SD slot.


Real numbers would be at least a minute.

 

[WizardStan]

It is just a regular SIGTERM, but you can also check the runlevel to see whether you're in normal operation or if it's about to be shutdown.

 

[_wb_]

*squints*


You want to reread what you just wrote? You just claimed that you can achieve 512+MB/s write speeds on Pandora's SD slot.


Real numbers would be at least a minute.

Oops, my bad, I was mentally confusing MB with KB, which of course doesn't help when you're computing a ballpark number . Obviously you're right, it would take quite a bit longer than a second.


With a class 10 SD card you should be able to get 10 MB/s write speed, so that would be about 51 seconds. Write speed is not really that important though, because who cares how fast it goes into suspend mode? Coming out of suspend is what needs to be fast, so there the read speed matters. Since the Pandora seems to be limited to about 18 MB/s, let's say it can be done in 30s. But of course in general you don't need to dump the entire ram: I assume you can ignore the caches and buffers and of course the free memory. Given that a lot of people comfortably use 256MB units, I assume that usually 180MB will be enough (10 seconds). If you only have the minimenu running, that number would be even lower.


That said, for me, just closing the lid is good enough

 

[notaz]

The CPU power consumption depends on its clock speed and its usage: I only have a Rebirth model so I don't know how the 1GHz model performs. The CPU seems to be the most efficient (per MHz) at 500 MHz: it uses about 0.7 W if used 100% at 500 MHz. At a lower clock speed it uses less power, e.g. about 0.5 W if used 100% at 350 MHz. At a higher clock speed it uses more power, e.g. about 1 W at 650 MHz, 1.5 W at 850 MHz. Note of course that in typical usage, the CPU will also be used less if you clock higher; these numbers are for 100% CPU usage.

What were you running while measuring this, hopefully not the stress test? That thing is a stress test, no typical programs are ever even close to using NEON that much, including PCSX or NEON optimized video decoders, it's just not practical. NEON increases power draw significantly, it even has it's own power domain (whatever that means).

While 0.11W sounds low at first, that is 110mW or 22 times more than my idle N900 consumes.

So you claim it only draws 1-2mA from the battery? That doesn't sound right, single LED draws ~5mA on Pandora.


That said there are some things that can't be properly turned off on pandora, like the nubs or USB host VBUS power supply. I complained about that to MWeston but it was too late, when I got power saving stuff working ED was already setting up rebirth production and trying to fix that would have resulted in months of delays. When the board is changed some prototypes need to be made to be sure changes are good, the bare board has delay over 1 month, then GC would need to reprogram machines just for a few test boards and so on (maybe ED can clarify), so it would be even harmful to do now, not to mention extra money needed for this.


Edit: it's unclear how much benefit those fixes would give though, nubs supposedly have their own automatic power saving modes and USB 5V supply in not loaded when nothing is connected.

 

[_wb_]

The CPU power consumption depends on its clock speed and its usage: I only have a Rebirth model so I don't know how the 1GHz model performs. The CPU seems to be the most efficient (per MHz) at 500 MHz: it uses about 0.7 W if used 100% at 500 MHz. At a lower clock speed it uses less power, e.g. about 0.5 W if used 100% at 350 MHz. At a higher clock speed it uses more power, e.g. about 1 W at 650 MHz, 1.5 W at 850 MHz. Note of course that in typical usage, the CPU will also be used less if you clock higher; these numbers are for 100% CPU usage.

What were you running while measuring this, hopefully not the stress test? That thing is a stress test, no typical programs are ever even close to using NEON that much, including PCSX or NEON optimized video decoders, it's just not practical. NEON increases power draw significantly, it even has it's own power domain (whatever that means).

Yes, the stress test. I know that's not realistic at all, but it does give a good upper bound. The estimate does seem to scale down quite well to more realistic scenarios.

While 0.11W sounds low at first, that is 110mW or 22 times more than my idle N900 consumes.

So you claim it only draws 1-2mA from the battery? That doesn't sound right, single LED draws ~5mA on Pandora.


That said there are some things that can't be properly turned off on pandora, like the nubs or USB host VBUS power supply. I complained about that to MWeston but it was too late, when I got power saving stuff working ED was already setting up rebirth production and trying to fix that would have resulted in months of delays. When the board is changed some prototypes need to be made to be sure changes are good, the bare board has delay over 1 month, then GC would need to reprogram machines just for a few test boards and so on (maybe ED can clarify), so it would be even harmful to do now, not to mention extra money needed for this.

I also have a hard time believing that an idle phone would use that little power, after all, it has to maintain a connection (or does that phone have some kind of "airplane mode"?)

 

[Exophase]

Radios and basebands have gone down in power a lot over the past few years. Waking it up every few seconds and checking to see if there's a signal uses very little power, and of course happens independently of the CPUs on the SoC (or independently of the SoC if it isn't integrated). It only has to do it ever few seconds because we're comfortable with it taking a few seconds of calling someone before their phone even starts ringing.


The power consumption of Pandora's wifi chip is pretty outrageous, much like it was for PSP and DS. I expect a big improvement in that category for Pandora 2.

 

[porg]

WiFi powersaving makes quite a difference !!!

 

I have tested the different power states.

Here are the absolute values (screen on with brightness 17, no BT, no USB, no TVOUT, sysinfo running and logging, WLAN in different modes)

 

Power mW WiFi state -- comment
- 945    WiFi off -- almost identical to WiFi on, powersaving on
- 950    WiFi on, powersaving on,  idle -- only occassional 1000-1600 spikes (beacons, handshakes & such I guess)
-1750    WiFi on, powersaving on,  active -- varies between 940-2300
-1750    WiFi on, powersaving off, idle -- almost identical to powersaving on active !
-2280    WiFi on, powersaving off, active -- quite a consumption

Sysinfo log attached:

wifi-power-states-commented--20121230-221908.txt

Note: Forum software does not permit file extension CSV. Change TXT to CSV after download.

 

 

 

[_wb_]

It's a good thing power saving is on by default. While active, the difference is not huge, but while idle, switching off the power saving basically doubles your power consumption.

Anyway, my Pandora's wifi is not that bad (I can get ~500 KB/s downloads), but compared to more recent wifi chips, it's not doing that well. With power saving on, the stability of connections can be quite weak at times (I really have to use mosh instead of ssh if I want to keep a session going). Also the range is noticeably weaker than on some other devices: when I'm two floors away from my wireless access point, it can just barely see its essid but it always fails to connect. My < 1 year old laptop can make a connection without any problems, my wife's smartphone can usually make the connection but it occasionally drops, and my iPad 1 can usually not make the connection but sometimes succeeds for a while.

 

[porg]

A big report resolving all mysteries about Pandora WiFi is coming up from me.

Data and insights already there. Just need the time to write up.

Will notify you.

UPDATE: Ok, here is my report.

@ _wb_ you read and commented it already anyhow. But for completeness, here is the link:

WiFi mysteries resolved once and for all — Power saving is clearly the culprit!

 

[danielo515]

Hey,

I'm very worried about power consumption.

I want to know if putting pandora in sleep mode ( moving the power button to the right  for a second) makes a difference with just closing the lid.

May in short periods of time does not make a difference, being sleep mode even worse?

I have 1GHZ unit, and I have it permanently at 1GHZ, because I think with this processor is the best speed - native one, isn't it?

May I underclock the CPU a bit?

Thanks in advance.

 

[WizardStan]

Underclocking doesn't have that big of an effect, the new power management system is brilliant and the system will use almost nothing when the CPU is idle.

Simply turning off the wifi and closing the lid seems to give me an idle time of about 60-70 hours from 100%, putting it into standby gives me about 200 hours of idle time. Anecdotally, I closed it with about 60% left Friday night, opened it again this morning (60 hours later) to find it still had about 30% left.