Power Consuption Approximation of Broadcom SOC


11 posts
by Spi » Tue Aug 28, 2012 8:43 am
Hey,
i use my RaspberryPi for the runtime analysis of a Kalman Filter on different hardware plattforms in my student research project.
For the compare of the used hardware plattforms i need a approximative value of the power consumption of the build in Broadcom BCM2835 SoC on the RasPi.

I used google but everything i got is the ARM Periferals "Datasheet", but it says nothing about power consumption. I found an approximation for the ARM1176JZF for different process geometries from TSMC, but i am not sure if any of it is used by Broadcom to maufacture the SoC.
Is somebody out there, who can give me a hint where i can get some informations about the power consumtion of this SoC (the Broadcom Homepage provides very few information) ?

For another ARM SoC from TI (Cortex A8) i got a power approximation spread sheet, something like this would be so great to have.
Posts: 2
Joined: Tue Aug 28, 2012 8:23 am
by RobertMM » Mon Sep 03, 2012 9:45 am
Hi,

Here are some numbers taken from Model-A and Model-B RPs.
Measurements were taken using one of these:
http://www.msoon.com/LabEquipment/PowerMonitor/

For these tests there was no display connected unless stated.

352mA - Model-B running stress test (for ARM)
330mA - Model-B idle
143mA - Model-A running stress test (for ARM)
118mA - Model-A idle (with HDMI connected)
110mA - Model-A idle
107mA - Model-B after "sudo halt"
106mA - Model-A, ARM_freq=300MHz (and for all below)
93mA - as above + SDRAM_freq=100MHz
93mA - Model-A idle, GPU_freq=100MHz
90mA - Model-A idle, GPU_freq=100MHz, Vdd=1.0V
80mA - Model-A idle, as above, + SD-card pulled
(power peaks as card is polled)
33mA - Model-A, after "sudo halt"

(you obviously need to multiple by 5 to get power in mW)

Hope that helps,
- Rob.
Posts: 2
Joined: Mon Sep 03, 2012 9:38 am
Location: Cambridge, UK
by Spi » Wed Sep 05, 2012 11:21 am
Hey,
the values are very helpful. Thanks for that.
Posts: 2
Joined: Tue Aug 28, 2012 8:23 am
by alias_neo » Fri Sep 07, 2012 11:27 am
RobertMM wrote:Hi,

Here are some numbers taken from Model-A and Model-B RPs.
Measurements were taken using one of these:
http://www.msoon.com/LabEquipment/PowerMonitor/

For these tests there was no display connected unless stated.

352mA - Model-B running stress test (for ARM)
330mA - Model-B idle
143mA - Model-A running stress test (for ARM)
118mA - Model-A idle (with HDMI connected)
110mA - Model-A idle
107mA - Model-B after "sudo halt"
106mA - Model-A, ARM_freq=300MHz (and for all below)
93mA - as above + SDRAM_freq=100MHz
93mA - Model-A idle, GPU_freq=100MHz
90mA - Model-A idle, GPU_freq=100MHz, Vdd=1.0V
80mA - Model-A idle, as above, + SD-card pulled
(power peaks as card is polled)
33mA - Model-A, after "sudo halt"

(you obviously need to multiple by 5 to get power in mW)

Hope that helps,
- Rob.



Can you elaborate as to what other hardware was connected?

I found that with HDMI, a USB wireless desktop adapter and ethernet, my 1A rated Samsung Galaxy S3 charger wouldn't cut it under load and would crash the system.
Posts: 103
Joined: Fri Jul 29, 2011 8:55 am
Location: Wilmslow, England
by asb » Fri Sep 07, 2012 11:31 am
No other hardware was connected other than a serial adapter, which did not affect power consumption.
Moderator
Posts: 756
Joined: Fri Sep 16, 2011 7:16 pm
by alias_neo » Fri Sep 07, 2012 11:48 am
asb wrote:No other hardware was connected other than a serial adapter, which did not affect power consumption.



Great, thanks. I need to get some measurements for the wireless desktop and ethernet.
Posts: 103
Joined: Fri Jul 29, 2011 8:55 am
Location: Wilmslow, England
by RobertMM » Fri Sep 07, 2012 4:53 pm
Numbers with USB devices connected:

(HDMI connected + 8GB Freecom USB memory stick)
450-490mA - Model-B sequential writes to USB drive
450mA - Model-B cat /dev/sda > /dev/null
400mA - Model-B idle

(HDMI connected + Netgear N150 USB wireless adapter)
480mA - Model-B running "apt-get update"
435mA - Model-B idle

(below: no HDMI and no USB devices unless otherwise stated)
352mA - Model-B running stress test (for ARM)
330mA - Model-B idle
143mA - Model-A running stress test (for ARM)
118mA - Model-A idle (with HDMI connected)
110mA - Model-A idle
107mA - Model-B after "sudo halt"
106mA - Model-A, ARM_freq=300MHz (and for all below)
93mA - as above + SDRAM_freq=100MHz
93mA - Model-A idle, GPU_freq=100MHz
90mA - Model-A idle, GPU_freq=100MHz, Vdd=1.0V
80mA - Model-A idle, as above, + SD-card pulled
(power peaks as card is polled)
33mA - Model-A, after "sudo halt"

- Rob.
Posts: 2
Joined: Mon Sep 03, 2012 9:38 am
Location: Cambridge, UK
by alias_neo » Sat Sep 08, 2012 8:34 am
RobertMM wrote:Numbers with USB devices connected:

(HDMI connected + 8GB Freecom USB memory stick)
450-490mA - Model-B sequential writes to USB drive
450mA - Model-B cat /dev/sda > /dev/null
400mA - Model-B idle

(HDMI connected + Netgear N150 USB wireless adapter)
480mA - Model-B running "apt-get update"
435mA - Model-B idle

(below: no HDMI and no USB devices unless otherwise stated)
352mA - Model-B running stress test (for ARM)
330mA - Model-B idle
143mA - Model-A running stress test (for ARM)
118mA - Model-A idle (with HDMI connected)
110mA - Model-A idle
107mA - Model-B after "sudo halt"
106mA - Model-A, ARM_freq=300MHz (and for all below)
93mA - as above + SDRAM_freq=100MHz
93mA - Model-A idle, GPU_freq=100MHz
90mA - Model-A idle, GPU_freq=100MHz, Vdd=1.0V
80mA - Model-A idle, as above, + SD-card pulled
(power peaks as card is polled)
33mA - Model-A, after "sudo halt"

- Rob.


This makes a little more sense and explains why I'm hitting near an AMP with X server running, ethernet, HDMI and wireless desktop adapter with Chromium browser making out the CPU and on 224 split.
Posts: 103
Joined: Fri Jul 29, 2011 8:55 am
Location: Wilmslow, England
by meltwater » Sat Sep 08, 2012 7:58 pm
Very useful measurements, thanks for doing them. Does the model A have the same linear 3.3 regulator? If so potential for even lower consumption with the hardware hack to use a switch mode one...

Don't suppose you have some bare metal measurements, wonder how close they come to the idle rates.

Thanks again!
______________
http://www.themagpi.com/
A Magazine for Raspberry Pi Users
Read Online or Download for Free.

My new book: goo.gl/dmVtsc

Meltwater's Pi Hardware - pihardware.com

Like the MagPi? @TheMagP1 @TheMagPiTeam
User avatar
Posts: 981
Joined: Tue Oct 18, 2011 11:38 am
by rew » Fri Nov 30, 2012 3:43 pm
meltwater wrote: Does the model A have the same linear 3.3 regulator? If so potential for even lower consumption with the hardware hack to use a switch mode one...

No, not much :-(

Most of the 3.3V went to the 9512 lan/usb chip. Which used a linear regulator to make it into 1.8V. Lots of wasted power there.

The Broadcom also needs a 1.2V powersupply. That however is generated from the 5V using a switch mode DCDC converter.

On the model B about half the 5V power went into the 3.3V regulator. Most of that went to the lan chip. So cutting off the lan chip reduces power by about half, but cuts the 3.3V usage by almost 90% (my estimate).
Check out our raspberry pi addons: http://www.bitwizard.nl/catalog/
User avatar
Posts: 391
Joined: Fri Aug 26, 2011 3:25 pm
by portets » Fri Nov 30, 2012 8:25 pm
rew wrote:The Broadcom also needs a 1.2V powersupply. That however is generated from the 5V using a switch mode DCDC converter.

The BCM has multiple +v pins. A 1.8v for itself and the ram that's stepped down using an internal switchmode. 3.3v powers various parts of it as well, like the hdmi. The 3.3v also powers the SD card. 2.5v linear powers the BCM's composite video out.

So replacing both the 3.3v and 1.8v linears with switching could be pretty beneficial.
Posts: 184
Joined: Sat Oct 29, 2011 6:24 am