13

I have one raspberry PI and two PI-2, side by side, inside a little cupboard.

I'm constantly monitoring their temperature. I'm worried because one of the Pi-2 is quite hot.

The PI and the hot PI-2 are running raspbian jessie and mostly idling.

The cool PI-2 is inside an official case (without active or passive cooling) and running OpenElec (latest stable).

I made some measurements on the Raspi1 and the hot Raspi2 a couple of months ago and everything seemed normal.

  • Raspi1 prev temp: ~50C (it was cased then)
  • HOT Raspi2 prev temp: ~ 40C

Now ambient temperature is quite hotter, but I wouldn't expect so much a difference:

  • Raspi1 current temp: ~45-50C (it isn't inside a case anymore)
  • Cool Raspi2 (cased) current temp: ~57C, seems pretty stable
  • HOT Raspi2 (not cased) current temp: ~63-75C, the CPU-GPU bundle is scalding to the touch (rest of the board doesn't feel too hot)

I know the theoretical limit is at 85C but I'm worried that it has jumped from a nice temp to something near the limit, while the cased one is notably colder.

All of them are connected via HDMI to a mostly unpowered TV, and through wired ethernet to the same router. I have interchanged power sources just in case, but didn't have any effect. I also tried to disconnect HDMI and network from the hot one, but no noticeable effect.

They are not overclocked, and anyway (as already said) mostly idling.

Software: In order to compare apples to apples, I swapped the SD cards and removed the case from the cool raspi2. The cool one has gone about 15C down, now sits about 42-43 (as the other one did a couple months ago). The hot one was down by 5C, but after a few minutes got hotter than it was (to over 65). So now I know it's not a software issue or something weird caused by the case. Ambient temp is about 30C, perhaps a little more.

To test both in as equal conditions as possible, I used the same power source and even the same ethernet cable.

To test, first I let the device idling for a while and look at the measure. Then I tried to max out the CPU, opening three ssh sessions and executing one instance of "while (true) do find /; done" and two of "while (true) do ls / -R; done", besides another session with top and another with temp measurements. top indicated a sustained CPU (i.e. top us% + sys%) usage of around 90%

The hot device was consistently at about 2.0 W when idling. When maxed out it was set to 3.0 W.

The cool one was at 1.5 W when idling. When maxed out it jumped between 2 and 2.5 W.

Power consumption: I finally got a usb voltage/current tester. According to it, the cool raspi2 is consuming ~0.4 A while running an idle openelec, while the hot raspi2 (same USB power source, same cable, same SD card) is consuming around 0.75 A. Voltage is around 4.90 V in both cases. ** This seems to be the problem. ** When trying to locate the issue, sometimes after fiddling for a while, temperatures went down to normal or near normal, to be up again after a couple of days.

What should I do now that I've isolated the issue?

raven
  • 254
  • 2
  • 10

1 Answers1

4

Based on the comment thread, I will put forward the following advice that has done well for me on both original RPi and RPi2: The quality of your power supply matters.

If the power supply you're using isn't adequately regulated or is trimmed too high, you may find your Pi is operating outside of its optimal electrical specifications. Running a power supply above 5.25 VDC can lead to current leakage inside the chip (even at idle) that can be measured as increased temperatures.

Not that I think anyone wants to read this, but here's a white paper on transistor current leakage and how ARM is designed to address that problem. It doesn't prove that this is the problem in this circumstance, but based on the comments, it certainly seems to point in that direction.

So, in short, quit buying cheap power supplies for your Pi. Not that good quality ones will be that expensive, but if you buy from eBay or (sometimes) AliExpress, you absolutely get what you pay for. Buying from a supplier that everyone would agree is reputable and concerned about customer satisfaction (Sparkfun and Adafruit being two easy examples, there are many others), you shouldn't have a problem.

WineSoaked
  • 350
  • 2
  • 10