RasPi power usage measurements ALL Models

[alexeames]

Got a Pi from RS in the post today

I've seen several power related threads where people are wondering how much the Pi uses and a few have attempted to make measurements. I've got a fairly accurate piece of kit I bought for measuring the current/voltage use of RC planes, so I thought I'd put it to use here.

Here's the setup.
Power source 3 cell lipo (12.5v)
Connected to Hyperion E-meter calibrate shunt (and then on to emeter)
Connected to LM2596S switching reg (allegedly 90% efficient) stepping voltage down to 5.0V
Connected to USB A port
Containing Samsung USB phone connection/charge lead
Connected to the Pi

Pi was connected to...
Network cable
HDMI monitor
Power in (described above)

No USB for now. I logged into the pi using SSH without attaching USB devices.

Raw data (needs converting later)...

On booting, the Pi drew 0.20 amps @ 12.5v
Once booted, it dropped to 0.19 amps @12.52v

While logging in via SSH, current went up to 0.20 A again
then fell back to 0.19 A.

Plugging in Logitech wireless dongle for kbd/mouse
current went from 0.19 A to 0.20 A and stayed there.

Typed startx on wireless kbd.
While loading 0.21 A
Back to 0.20 A on idle (all @ 12.5V)

Start midori 0.21A while loading
Back to 0.20 Aonce loaded.

Load Raspi forum main page
0.21 A fell back to 0.20 once loaded

Load Raspi site main page
0.21 A fell back to 0.20 once loaded
then fell back to 0.19 on idle

Then shut down. 0.21 A while shutting down.

So what does this tell us?

1) Assuming the 90% efficiency is true, the idle power usage with just network and HDMI monitor connected is 0.9 * 12.5 * 0.19 = 2.14 watts

2) My logitech wireless kbd/mouse dongle required only 0.9 * 12.5 * 0.01 = 0.11 watts, which is nice and low

3) loading X or a web page or a program use a similar amount of power as 2) ~ 0.11 watts

I haven't had a chance to stress the unit yet, but will make further measurements when I do. I'm fairly confident of the accuracy of the measurements, but less so about the efficiency (90%) of the reg. But if it's less efficient, that means the pi uses less.

Bottom line. Pi uses about 2 watts at idle.

Stressed Pi measurements to come later. I've got a fairly inefficient python anagram algorithm I wrote a few weeks ago that gobbles up processor power. And I might hook up the USB data logging facility on the emeter if I can remember how it works to give you some pretty graphs and stuff.

All good fun. Hope this is useful for someone, but either way, I'm having fun.

[alexeames]

I did the anagram thing. Inefficient anagram finding algorithm using multiple nested loops and a huge word list. It takes my main quad core i7 18 seconds to search a 178,000 word file and find all the anagrams for dogs. It works the processor quite hard (and the fan goes to max). It took my Pi 8 minutes and 16 seconds. But the difference between idle and under load isn't much. It's the same as it was before when loading a web page etc...

Here's a pretty chart (I did the data logging thing).


The graph shows a ~0.1 watt difference between idle and under cpu load. There are a few stray readings, but we're working at the precision limit of the equipment (and I guess cpu load is not always constant either).

I wonder how to hammer the GPU?

(By the way, I now have a lightning fast anagram finder, but deliberately used the inefficient one to tax the processor. It was all a python learning exercise.)

[tufty]

I wonder how to hammer the GPU?

omxplayer? quake?

[prodata]

Got a Pi from RS in the post today
Connected to LM2596S switching reg (allegedly 90% efficient) stepping voltage down to 5.0V

Sorry but I'd be fairly sceptical of the 90% figure - that's probably the maximum with all of the I/O voltages and current at their most favourable values. With a 5v output I'd suspect it drops to 80% max and even that would only be at the design current (which I think might be 3A). Efficiency often isn't linear for major differences in current draw and so for 0.4A it could be down in the 60-70% range and maybe even lower, eg towards 50%.

So I'm sure you're in the right ballpark but I'd guess very likely still overestimating the draw quite significantly.

0.1ohm 1% resistors are easy to obtain and if connected in series with a 5v supply and checked with a high impedance millivoltmeter ought to give a good indication of actual current draw. eg 200mA would show a 20mV voltage drop across the resistor while having a negligible effect on the supply voltage.

[alexeames]

Got a Pi from RS in the post today
Connected to LM2596S switching reg (allegedly 90% efficient) stepping voltage down to 5.0V

Sorry but I'd be fairly sceptical of the 90% figure - that's probably the maximum with all of the I/O voltages and current at their most favourable values. With a 5v output I'd suspect it drops to 80% max and even that would only be at the design current (which I think might be 3A). Efficiency often isn't linear for major differences in current draw and so for 0.4A it could be down in the 60-70% range and maybe even lower, eg towards 50%.

Don't be sorry, I'm sceptical about it too, which is why I flagged it up. It looks to me as if we might be able to run this thing on 1 watt. Gert already said if a switching reg was swapped in for the linear one on the Pi board it would greatly increase efficiency. Doesn't look all that easy to do (to me) though if it's RG2 on the board.

If I splice up a couple of leads I could proably measure the actual power into the Pi instead of between battery and reg. That would give a true Pi consumption figure, but in reality you do normally have to use some kind of a reg (and I'm not sure i want to destroy a USB lead)

[WildWalker]

I have mine plugged in to a mains power display.

I have two of these from different manufacturers, and they both report around 4-5 Watts (AC 240v) even when the Pi is working.

Alan.

[prodata]

I have mine plugged in to a mains power display.

I have two of these from different manufacturers, and they both report around 4-5 Watts (AC 240v) even when the Pi is working.

But again if you're referring to a typical 'Kill-a-Watt' meter plugged into the mains socket then this will be measuring the total power draw, including all the inefficiency in the 5V adapter. It's probably also right at the low end of what the meter can measure and so potentially inaccurate for that reason too. So it gives you an idea of what the maximum power draw at the mains socket might be, certainly, which is in the expected range. But it's still not giving an accurate fix on the power draw by the Pi itself on the 5v line, which is the information of real interest for planning off-grid applications, eg running from batteries.

[alexeames]

But it's still not giving an accurate fix on the power draw by the Pi itself on the 5v line, which is the information of real interest for planning off-grid applications, eg running from batteries.

Are people going to be running from batteries without a regulator? To be honest, I think the more pertinent info is WITH regulator. Although I am curious to see what the efficiency of my reg ACTUALLY is, so I will probably patch something up and take measurements later on. I've found I've already got the necessary bits to make an adaptor instead of destroying a USB lead.

I will repeat yesterdays experiment with the measurement between the reg and the Pi. Tune in later for another exciting episode of...


Measure your Pi's Power

[AndrewS]

I'd expect a battery (12VDC ?) -> 5V regulator to be much more efficient than a mains (240VAC) -> 5V regulator

[Gert van Loo]

Gert already said if a switching reg was swapped in for the linear one on the Pi board it would greatly increase efficiency. Doesn't look all that easy to do (to me) though if it's RG2 on the board.

I have been thinking about a solution for the power wastage. But I have been busy with two other boards (Gertboard is one of them. I expect an announcement from Liz any time about the other).
Any idea how many people would want a minimum-power-wasted solution even if they have to remove RG1 and RG2 from their board?

[AndrewS]

Is the entire heatsink-area under RG2 all soldered? Does that make it extra hard to remove? Or mean I'm more likely to damage other components while trying to remove it?
How big an estimated power-saving would be gained by replacing RG1 and RG2 with more efficient versions?

[prodata]

Are people going to be running from batteries without a regulator? To be honest, I think the more pertinent info is WITH regulator.

Don't disagree at all. The only point I'd make is that the regulators are going to be another variable and it would be interesting to have the actual Pi draw as a benchmark or baseline. And then the effect of any regulator can be added in on top of that. I don't know what sort of (presumably USB) mains adapter WildWalker was using but it's not spelt out whether it was even a switching one, let alone a high efficiency one. So while 4W sounds reasonably low, the actual Pi draw could have been 2W or even lower, eg down towards 1W on idle, as one might perhaps expect.

[alexeames]

Are people going to be running from batteries without a regulator? To be honest, I think the more pertinent info is WITH regulator.

Don't disagree at all. The only point I'd make is that the regulators are going to be another variable and it would be interesting to have the actual Pi draw as a benchmark or baseline. And then the effect of any regulator can be added in on top of that. I don't know what sort of (presumably USB) mains adapter WildWalker was using but it's not spelt out whether it was even a switching one, let alone a high efficiency one. So while 4W sounds reasonably low, the actual Pi draw could have been 2W or even lower, eg down towards 1W on idle, as one might perhaps expect.

That would certainly agree with my figures from yesterday if your guess about the Reg efficiency is correct (which it probably is). Adaptor is made. Off to do some measurements

[Gert van Loo]

Is the entire heatsink-area under RG2 all soldered? Does that make it extra hard to remove? Or mean I'm more likely to damage other components while trying to remove it?
How big an estimated power-saving would be gained by replacing RG1 and RG2 with more efficient versions?

My plan was to try it on some boards when a few more are available. Then post a video. You will need a good hot-air gun (Not paint stripper or hair dryer!) and I was thinking of using tin foil around the area to protect the surrounding components from the heat.
I think it could half the power consumption in some cases. The idea is that for battery powered usage even saving 10% is a lot.

[AndrewS]

You will need a good hot-air gun

I've got a friend with one of those. He used it to fix my PS3 by reflowing the BGAs (fixed long enough for me to recover my game-saves at least)

I think it could half the power consumption in some cases. The idea is that for battery powered usage even saving 10% is a lot.

Wow! Definitely significant...

[alexeames]

New test. Got to go pickup kids at 1430 so no time for full data yet.

Measuring exactly the same setup as in post 1 except calibrated shunt placed between reg and Pi. Running with HDMI monitor and network cable, connected via SSH. Ran the same anagram program on the same word. It took ~8 minutes. Left it a minute to stabilise after it finished, then sudo shutdown -h now.

After boot, it stabilised on 1.95 watts
While being hammered by anagram prog it flickered between 2.09 & 2.14 watts
When that finished it stabilised on 1.99 watts
On shutdown it went up to 2.14 watts and stayed there even when shut down (don't quite understand that)

Pretty graph later. If anybody wants the volt/amp/time data I can make that available.

The measured currents were obviously higher, so more "in range" than yesterday and the voltages obviously lower.

Bottom line. Pi uses just under 2 watts when idle and just over 2 watts when performing a long anagram search.

[poing]

Is the entire heatsink-area under RG2 all soldered? Does that make it extra hard to remove? Or mean I'm more likely to damage other components while trying to remove it?
How big an estimated power-saving would be gained by replacing RG1 and RG2 with more efficient versions?

My plan was to try it on some boards when a few more are available. Then post a video. You will need a good hot-air gun (Not paint stripper or hair dryer!) and I was thinking of using tin foil around the area to protect the surrounding components from the heat.
I think it could half the power consumption in some cases. The idea is that for battery powered usage even saving 10% is a lot.

Gert already said if a switching reg was swapped in for the linear one on the Pi board it would greatly increase efficiency. Doesn't look all that easy to do (to me) though if it's RG2 on the board.

I have been thinking about a solution for the power wastage. But I have been busy with two other boards (Gertboard is one of them. I expect an announcement from Liz any time about the other).
Any idea how many people would want a minimum-power-wasted solution even if they have to remove RG1 and RG2 from their board?

Reading a post from you a while ago I planned to ask you to do this when I would get my second Pi, although I probably have to wait until Christmas to receive my first one (OK, I only ordered March 7, so I'm not complaining)

Most things I want to do with the Pi seem to be battery powered, as in portable embedded devices. 10% power saving would be a lot, something like 40% could mean the difference between usable and unusable. So I truly hope you will post a detailed description on how to squeeze the most out of a Pi running from batteries.

[alexeames]

Graphs now done...

[Gert van Loo]

[quote="poing"
Reading a post from you a while ago I planned to ask you to do this when I would get my second Pi, although I probably have to wait until Christmas to receive my first one (OK, I only ordered March 7, so I'm not complaining)
[/quote]

There is only so much I can do and this one is rather far down the list because there are not that many users who are willing or capably of making a modification which is so complex and risky.

I would also like to point out that working with SW is in principle free, but developing hardware requires money to get PCBs made, components ordered and sometimes pay a technician to solder them. (I can't do 0603 let alone 0402 parts)

[alexeames]

There is only so much I can do and this one is rather far down the list because there are not that many users who are willing or capably of making a modification which is so complex and risky.

Too right. When I looked at RG2 and RG1, I thought. Hmmm. How badly do I need a switching reg on this? I don't have that sort of heat gun But someone on here must have one. (Not sure who'd want to risk wrecking other people's hardware though).

I would also like to point out that working with SW is in principle free, but developing hardware requires money to get PCBs made, components ordered and sometimes pay a technician to solder them.

Quite right. So let's not be asking Gert to mod our boards for us guys. He's doing loads of good stuff already. Thanks Gert.

[prodata]

I suspect this may be more in the realm of needing to wait until a low-power version of the Pi might be considered for manufacture. It does look like 2W may be the minimum continuous power consumption, which is still actually quite high for a unit needing to run, for example, 24/7 from solar power in the UK. If 1W were a real possibility then that could halve the necessary investment in solar panel/battery etc, potentially saving eg £200-300 for an installation.

There is a software angle to this however. Various embedded-type applications don't necessarily need literally continuous computation. The aims could sometimes be achieved with a computer that was asleep (ie very low power mode but still minimally functional) but that woke up every eg 1/5/60/whatever minutes to perform some task and then reverted to sleep mode for its next 'long' sleep. Anyone know whether this would be achievable with the Pi hardware and, if so, how would you go about it - at least in block diagram terms to start with (ie what interactions of hardware/driver/Debian/application code might be involved)? Any notion at all as to how much power saving might be available from this approach?

[jbeale]

Re: removing RG1, RG2: with a suitably powerful soldering iron it can be done quickly even though the copper surround draws heat away. Often easier if you actually add solder at first, to help the heat transfer. If you have a small soldering iron with a small tip, it might be difficult to impossible. There is a low-melting-point solder product called ChipQuick made for this very purpose, flows easily around pins and helps transfer heat, and actually alloys with the existing solder and lowers its melting point (maybe too expensive for casual use though.) At any rate, that project is definitely not in the "absolute beginners" category.

Mode switching to a low-power state is a very good approach if possible, but I gather the SoC wasn't designed with this in mind(?) And the bootup process seems to take the better part of a minute, so the full on/off cycle seems justified only if you are "on" for a substantial amount of time.

[alexeames]

Someone suggested trying oxmplayer to hammer the GPU. So I thought I'd have a go at that. My son was underwhelmed on day 1, so I thought it would be a good idea if I set him straight by making the Pi play some HD vid. (He has decided it's much cooler now.)

Similar setup to before. No pc logging or graphs, this time, just me noting down some amp and volt readings at various times. HDMI, Ethernet, Logitech USB kbd/mouse dongle, USB stick, via shortish extension cable, power in from 3 cell lipo with reg. Calibrated shunt connected between regulator and Pi.

While booting 0.48A @ 4.98v = 2.39 watts (slightly more stuff plugged in today)

stabilised after boot 0.44A @ 4.98v = 2.19 watts *

Started omxplayer 1080p Big Buck Bunny demo Full HD film with sound via HDMI
Current 0.52A-0.54 @ 4.98v .51 lowest .55 highest. Voltage was rock steady.
Call it 0.53A @ 4.98v = 2.64 watts

cancelled video after a couple of minutes (we'd already watched it)
reverted to 0.44A @ 4.98v = 2.19 watts
shutdown -h now 0.47A @ 4.98v = 2.34 watts

So the difference between idle and running a 1080p full HD vid is 2.64 - 2.19 = 0.45 watts
So now we know That's the hardest I've managed to push my Pi so far and we're at 2.64 watts.
Bung in the extra 15% the 5v regulator eats and we're at 2.64/0.85 = 3.15 watts that actually comes out the battery.

___
As an aside, I noticed yesterday that the current consumption goes up slightly when the Pi is shut down.
Happened again. It's reproducible. Anyone know why that is?

*Baseline numbers slightly higher today because I had two USB devices plugged in.

[jojopi]

You will need a good hot-air gun (Not paint stripper or hair dryer!) and I was thinking of using tin foil around the area to protect the surrounding components from the heat.
I think it could half the power consumption in some cases. The idea is that for battery powered usage even saving 10% is a lot.

Are there any drop-in replacement switch-mode regulator ICs? I was assuming not; that it would require both ICs and inductors at least. In that case there is no need to remove the original LDOs. Just cut their legs off and hotglue a daughterboard to the carcass. Much safer for non-experts.

Of course, until availability normalises everyone should avoid voiding their warranties whichever way.

[AndrewS]

As an aside, I noticed yesterday that the current consumption goes up slightly when the Pi is shut down.
Happened again. It's reproducible. Anyone know why that is?

Possibly all the write buffers getting flushed to disk? (where in this case "disk" is the SD card).
And/or the power-management features of the SoC being disabled?