RasPi power usage measurements ALL Models

[alexeames]

With the new firmware I measured 0.107 A after shutdown -h now.
Then I waited 15 for the temperature to re-equilibriate and took this picture:

Looks like the LAN chip is still drawing quite a bit of current.

Excellent. Thanks for posting that. Looks like the LAN chip and the reg RG2 are the guilty parties. Glad your current measurement is very close to mine. Three within 10% of each other gives us confidence in the figure.

[dom]

Looks like the LAN chip is still drawing quite a bit of current.

Thanks for this.
I believe the LAN chip is held in reset which I would expect to minimise power. Any suggestions?

[azerty]

Looks like the LAN chip is still drawing quite a bit of current.

Thanks for this.
I believe the LAN chip is held in reset which I would expect to minimise power. Any suggestions?

In the LAN9512 datasheet it says that the default USB suspend state reduces current down to 4.2 mA typical, so my first guess would be to find some way to drop the 9512 to suspend mode before powering off. Is that even possible? Linux power management confuses me

[AndrewS]

But if you're plugged into an HDMI TV, there's a good chance you're running off mains power, in which case a deep-sleep mode isn't so relevant. I'm thinking battery-powered situations...

If no HDMI TV, then that pin is available and reasonably accessible. Or you'd prefer one of the GPIO header pins?

Interesting point. But I'd much prefer one of the GPIO header pins (if this feature ever gets added) - we clearly have different definitions of "reasonably accessible"

[Alligator]

I'd second a GPIO pin as a way to both trigger and wake from sleep. This opens up a lot of possibilities.

[Wolfram23]

I'm really curious where this is going. Will the RPi have an auto sleep/wake function based on a GPIO input? Would be extremely helpful for a Carputer scenario, although I suppose a secondary logic controller like, perhaps and Arduino?, could do the job. I really don't know enough to know what is and is not possible right now.

[pluggy]

I measured my Pi power consumption and it comes out at ~2 watts with just ethernet connected, sometimes a little under sometimes a little over. My 'shunt' was a home made wire wound 0.1R resister, I'm not 100% confident of its exact value so I'm not going into too much detail with it. But being in the same ball park makes me think Its not too far away. I also measured the CPU/RAM & USB chips with an IR thermometer, both in the high forties C after several days running and a steady but not excessive network traffic load. The USB was a couple of degrees higher than the CPU. The Atmega 328 (ala Arduino) I had hooked up to the GPIO serial port runs around 28C . Ambient was around 21C.

[Alligator]

The real question here is how can we sleep and wake up the Pi to save even more power. Many of us are waiting for an answer from someone more knowledgeable than ourselves. For now, I'm planning to run my Pi off of a battery with a 12 volt timer and dc-dc converter. This will at least let me turn it on and off. With a cron job, I can shut it down automatically and wait for the timer to turn it off. However, there should be something easier than this!

[alexeames]

I measured my Pi power consumption and it comes out at ~2 watts with just ethernet connected, sometimes a little under sometimes a little over. My 'shunt' was a home made wire wound 0.1R resister, I'm not 100% confident of its exact value so I'm not going into too much detail with it. But being in the same ball park makes me think Its not too far away.

Sounds about right to me.

I also measured the CPU/RAM & USB chips with an IR thermometer, both in the high forties C after several days running and a steady but not excessive network traffic load. The USB was a couple of degrees higher than the CPU. The Atmega 328 (ala Arduino) I had hooked up to the GPIO serial port runs around 28C . Ambient was around 21C.

That's interesting. Some people are adding heat-sinks (jewellery ), but according to Broadcom folk, the chips are fine even when running in a really hot environment.

[AndrewS]

For now, I'm planning to run my Pi off of a battery with a 12 volt timer and dc-dc converter. This will at least let me turn it on and off. With a cron job, I can shut it down automatically and wait for the timer to turn it off. However, there should be something easier than this!

That's fine, as long as you have some way of ensuring the correct time is set on the RPi after each bootup (e.g. an RTC attached via i2c, or a permanent internet connection you can use for NTP)

[RolfBly]

Impressive work being done here.

Some thoughts on solar powered applications. I think it's safe to say current draw in idle mode is 100..120 mA, and it's not just the processor that eats it but also other components. So I guess it'll be hard to drive that number down with firmware changes. And I think hardware changes with SMD's are a no-go, because you'd need specialised, expensive, eco-unfriendly equipment. So, please correct me if I'm wrong, 100..120 mA in idle mode is as good as it gets.

In solar powered application you normally do tasks x, y, z and then go hibernate for some amount of time, wake up, do x, y, z, perhaps compute the next hibernation interval, hibernate, and so on.

In hibernation, you want power consumption to be close to zero. 120 mA (~0.5 W) could still be way too much in many situations.
So you'd need an 'independent' external device that reads and stores that hibernation interval, shuts power to the Pi off, turns it on when the time comes, waits for the next interval to read, and so on.
You could do it with a low power relay and perhaps an ATtiny13A, which draws only 24 uA (microAmps ) in idle mode, 190 uA in active mode.

[alexeames]

azerty has done a regulator swap for switching regs. Details here, but it saves 27%, which is not as much as hoped.

http://www.raspberrypi.org/phpBB3/viewt ... 63&t=12387

The conclusion, corroborated by the IR photos from a few posts back, was the LAN chip is a power hog.

[jamodio]

Did somebody check how much current is being drawn from the RG1 (1.8V) regulator ?

There is a connection that shouldn't be there from the LAN9512 VDD18CORE pins to the 1V8 rail, those two pins are not an input for the 1.8V LAN9512 core, they are there for the decoupling caps for the internal 1.8V regulator and are not supposed to be used to feed any other parts.

Hard to say without removing stuff if some current is being drawn from those pins to the 1V8 plane, if so that could contribute to additional heat and current draw from the LAN9512.

I also found that the LDO voltage regulators populated on the production boards are a cheap Chinese knockoff of the original ON Semi used on the beta boards. The cheap part seems to have a higher drift in output voltage in function of the ambient temperature.

While apparently the temperature of the LAN9512 and SoC (unfortunately electrical and other specs are not publicly available,) are within spec, adding heatsinks provides more area for heat dissipation and reduces the heat transfer through the printed circuit board.

Using DC-DC switched converters not only provides more power efficiency but reduces the dissipated heat particularly if the board is inside a case.

What are the short/medium/long term effects, hard to say and I believe we are not doing much to have sort of a detailed ticketing system to track reported failures and fixes.

-J

[Wolfram23]

The RPi has a decent sleep mode at 120mA or so, but is there any way to enable Hibernate so that it unloads all RAM to the SD card and completely shuts down? Would be nice to have that for faster boot times and lower power.

[AndrewS]

The RPi has a decent sleep mode at 120mA or so, but is there any way to enable Hibernate so that it unloads all RAM to the SD card and completely shuts down? Would be nice to have that for faster boot times and lower power.

I believe hibernate should be possible with the right software (although AFAIK no-one's done it yet) https://www.google.co.uk/search?q=linux+hibernate
But without adding some kind of external power-control board, the Raspi has no way to "switch itself off" - the regulators are always connected to the MicroUSB power input. Rurwin had an idea for such a board just using two relays http://www.raspberrypi.org/phpBB3/viewt ... 02#p104402

[alexeames]

The RPi has a decent sleep mode at 120mA or so, but is there any way to enable Hibernate so that it unloads all RAM to the SD card and completely shuts down? Would be nice to have that for faster boot times and lower power.

I believe hibernate should be possible with the right software (although AFAIK no-one's done it yet) https://www.google.co.uk/search?q=linux+hibernate
But without adding some kind of external power-control board, the Raspi has no way to "switch itself off" - the regulators are always connected to the MicroUSB power input. Rurwin had an idea for such a board just using two relays http://www.raspberrypi.org/phpBB3/viewt ... 02#p104402

It's all a bit above my paygrade, but wasn't the outcome of that thread that in theory it would be possible with relays, but in practice relays with such a low current consumption were not available and it should be done with other active components?

[Wolfram23]

Well, the DC-DC buck chip I'm looking at has a low power mode based on whether one of it's inputs is high or low, so I think there might be a decent solution to my carPi issue, at any rate. Run the 12V ignition (switched) line to a buck chip to get 3.3V. Then connect it to:
1) GPIO pin that will auto enable shut down sequence
2) Timer that will run for however long (30 seconds?) and is connected to the 12V Batt line (constant power) buck chip, and "turn it off".

The effect would be, once the car is turned off, the Pi will auto shut down and then all the circuitry will also turn off. Turning on car power (whether "2 clicks" of the key or turning the car on) should then cause the 3.3V buck chip to turn on the 5V chip, which would then boot the RPi as soon as it has power.

Now, if I can get auto-hybernate to enable instead of "sleep" on the RPi, that would be extra nice. Really wish I had my Pi already... 11 weeks to go

[AndrewS]

Now, if I can get auto-hybernate to enable instead of "sleep" on the RPi, that would be extra nice.

I think that hibernation on RPi (with the limited write-speed on the SD card) might be slower than a "normal shutdown"? I guess there's plenty of benchmarking to be done if/when somebody gets hibernation on the Pi working

[Wolfram23]

Yeah it probably would be slower, but the main reason isn't to speed up shut down, but boot up. Then again, maybe the basic startup would be ok, I don't actually know yet. Ideally though I want the RPi to turn on and start playing music in less than 20 seconds.

[RolfBly]

@AlexEames: There are bistable a.k.a. latching relais. You give them a pulse, they toggle. So you'd only need a fairly beefy current at switching time, not continuously.

if I only had time

[danpeirce]

With the new firmware I measured 0.107 A after shutdown -h now.
Then I waited 15 for the temperature to re-equilibriate and took this picture:

Looks like the LAN chip is still drawing quite a bit of current.

Nice image.
That does seem like a great deal of power once the system has halted but I am guessing this was not actually something that was a priority when the board was designed. I'm thinking the priorities were to keep the cost down while still getting some performance when the system is running.

[Rabjerg]

So, i was just bored, and thought of getting some measurements, do what you want with them, and say if you want some more or some other.
I used a power supply with USB interface and logging software.

Following setup.

Fresh Debian Wheezy Image, with all updates installed before logging.
HDMI monitor Connected.
LAN Connected.
Logitech Wireless Keyboard and mouse Dongle on USB-0. (MK250)

Power supplied via GPIO pins, bypassing the fuse.
Current limited to 1000mA.

Time span, 13min 00sec
Logging interval: 2000ms (2sec)

Values measured.

Supply voltage: 4,900 (V)
Maximum value: 4,907
Minimum value: 4,894
Range: 0,013
Average: 4,897

Current (A):
Maximum value: 0,51
Minimum value: 0,05
Range: 0,46
Average: 0,403

Power (W):
Maximum value: 2,497
Minimum value: 0,245
Range: 2,253
Average: 1,976

During boot, the power usage was between 1,92W and 2,30W (Spike 2,49W)
Idle terminal (eg. before starting X), between 1,92W and 1,96W
Starting X and idle X, 1,96W to 2,02W
Starting Task manager and a Pygame (Squirrel eats squirrel, 100% CPU load, GPU load unknown), 1,98W to 2,16W (Spike 2,49W)
Running a Pygame, 2,12W to 2,16W.
Shutting down, 2,12W to 2,40W
Halt, 0,47W

[ivanohé]

I've read that new model uses 0.12W when halted. That's great, because I was looking for a remote controlled switch.
So it would be completely usefull if the pi could be awaken from halt state, is it possible ?

[RaTTuS]

I've read that new model uses 0.12W when halted. That's great, because I was looking for a remote controlled switch.
So it would be completely usefull if the pi could be awaken from halt state, is it possible ?

only by power cycling it .....

[ninadpchaudhari]

Damn !!
I never knew RPi draws such low Currents !!
I made a power supply with transformers , 4 LM05 s which still gives me 4.35 v

Though the ckt. ideally can take upto 2A !!