Thermal comparison of three embedded Linux boards
Thermal imaging a Raspberry Pi 1 and 2, and BeagleBone Black, idle and under load.
Before sending back a Flir E5 camera that we have on loan, I thought I'd compare the peak temperature of a Raspberry Pi 1 Model B, Pi 2 Model B and BeagleBone Black, when each of these are idle and again when they are under load. I should point out that this is not a particularly scientific test, but rather a matter of curiosity and seeing how each fares in terms of generating heat.
To start with an image was taken with each board cold.
First up was the single core Raspberry Pi 1 Model B.
Followed by the quad-core Pi 2 Model B.
And finally, the BeagleBone Black.
As one would hope, only minor variation can be seen in the peak temperatures — displayed top-right of each image — and the spot temperature located roughly in the centre of each board. And even then these may be down to the varying emissivity/reflectivity of each board.
The boards were then imaged again after 30 minutes of being powered-up and with no load.
Here we can see that part of the Raspberry Pi 1 Model B has risen to 47.6C.
Whereas the Pi 2 after the same period of time has only peaked at 41C. This is most likely a side effect of the power supply being upgraded from using a linear to a switching voltage regulator, with the former being much more wasteful and effectively throwing away energy as heat.
The BeagleBone Black came in a little hotter after 30 minutes, peaking at 43.5C. Although different SoCs and boards integrate different peripherals, and this will contribute to the total power dissipation, along with the number of ARM cores and their clock speed etc.
Next the boards were put under load using the stress-ng tool, with:
$ stress-ng -a 1
This can be run with many different options and if you really want to put a system to test, these should be tuned for that particular system and perhaps also the subsystem that you are most interested in. However, since we just wanted to generate a decent load and that was the same on each board, specifying “-a 1” should suffice as it runs one worker (instance) of every test.
After 30 minutes under stress we can see that the Raspberry Pi Model B reached 53.6C.
Whereas the Pi 2 got that little bit hotter at 56.7C after 30 minutes under load.
And the BeagleBone Black peaked at 54.6C.
Interpreting the results
First, some caveats:
Peak temperatures were observed and not total heat generated
A workbench is hardly an environmental test chamber!
Obligatory disclaimers aside, the results are pretty much as you would hope, with the Pi 2 running cooler than the Pi 1 when idle, more than likely thanks to an improved power supply. Under stress the single-core Pi and BBB peaked at around the same temperature, whereas the Pi 2 got hotter, but it was almost certainly getting more done when all those various stress tests were spun up.
If we wanted to accurately measure the comparative thermal dissipation, and perhaps make more sense of this by also measuring the compute performance and efficiency, we'd need to be a lot more methodical and to have a much more controlled test environment. However, these simple tests have provided an insight into the comparative temperatures you might expect to see with each board when they're put under stress.
Get your hands on a Flir E5 Thermal Imaging Camera
Thermal imaging can be very useful for identifying changes in temperature. We decided to put this to the test and undertook a study to understand what happens to our bodies when we consume chillies. Staggeringly, the body temperature of our guinea pig increased by 4 degrees!
Open source (hardware and software!) advocate, Treasurer and Director of the Free and Open Source Silicon Foundation, organiser of Wuthering Bytes technology festival and founder of the Open Source Hardware User Group.
May 12, 2015 16:53
CommentsAdd a comment
December 28, 2015 11:32
Thought I'd mention Seek Thermal as a great and budget-friendly alternative to Flir. They have iPhone and Android versions as well as a new standalone version. The iPhone/Android one is cheaper than the Flir One and higher resolution. The software isn't quite as nice maybe, but gets the job done.
I don't know if links are allowed here, so I won't link it (not spam and I'm not affiliated with Seek Thermal, just an owner/user).
May 19, 2015 12:01
If you pass a constant current through a resistor, you will produce a voltage across it. An electric potential difference between the two ends.
As I said, a thrmal circuit is LIKE an electric circuit. If you generate a constant heat (in Watts) in a device with a constant Thermal resistance (not expanding, etc) you will produce a thermal difference between the inside junction (silicon to case) and the outside junction (case to air).
You use this principle every time you add a heatsink: the "case to air" resistance, quoted in degrees C per watt, is lowered. A given number of watts produce a smaller temperature rise. The actual number of Watts PRODUCED remains a function of device voltage and current.
Hence the need to measure voltage and current draw under load, as well as temperature.
May 18, 2015 14:19
"Thermal circuits are like electric circuits - but the current is heat, and voltage is temperature." Sorry but wrong Heat is energy .. The combination of Current and Voltage is energy measured in watts can be converted to BTU's. if current or voltage is zero no energy used no work done you need both . Temperature is the result of work done in the environment it is in
May 18, 2015 12:32
> Thermal circuits are like electric circuits - but the current is heat, and voltage
> is temperature.
> Adding a simple heatsink lowers thermal resistance, so a given amount of heat creates
> a lower temperature difference.
> So the best thing to in a test like this is : add a current meter to the power supply
> line. That gives you actual power being dissated, in watts.
> The thermal images just tell you the best places to add heatsinks.
May 18, 2015 11:11
Its interesting to note that at the distance the images have been taken, the visual part and the thermal parts of the image combined by their MSX technology are not well aligned. If you look at the images, the hot areas are actually above the location of any components. In the case of the Pi1 Idle the hottest areas are shown on the FLIR as being the Ethernet connector and the bottom of the HDMI socket. Using a different thermal imager (in this case a Land Instruments ARC) with just a pure thermal image, its easier to see that the hot spot are actually the Ethernet PHY and the power regulator.
May 18, 2015 10:47
Thermal circuits are like electric circuits - but the current is heat, and voltage is temperature.
Adding a simple heatsink lowers thermal resistance, so a given amount of heat creates a lower temperature difference.
So the best thing to in a test like this is : add a current meter to the power supply line. That gives you actual power being dissated, in watts.
The thermal images just tell you the best places to add heatsinks.