Pi 3 Overheating Issue?

[ItsMurf]

I read this article on Make: today that stated, that the Pi 3 was getting to temperatures as high as 87.6 Celsius! I don't have my Pi 3 yet does anyone have this problem?


"Make: Article" http://makezine.com/2016/03/02/raspberr ... atch-fire/

[chrisoh]

There's a thread here about this with some more info and commentary: viewtopic.php?f=63&t=138193

[DougieLawson]

I read this article on Make: today that stated, that the Pi 3 was getting to temperatures as high as 87.6 Celsius! I don't have my Pi 3 yet does anyone have this problem?

It's not a problem and my Pi3 didn't catch fire when I got it to 82.5°C - the big make -j4 compile I was running completed OK.

[W. H. Heydt]

I read this article on Make: today that stated, that the Pi 3 was getting to temperatures as high as 87.6 Celsius! I don't have my Pi 3 yet does anyone have this problem?


"Make: Article" http://makezine.com/2016/03/02/raspberr ... atch-fire/

The article title appears to be from an old programming joke list of suggested additional machine instructions. I was always kind of partial to "Backspace and stretch tape" and "Execute operator immediate".

[oryxhasapi]

if you are worried get a heatsink

[gordon77]

if you are worried get a heatsink

or fit one of these http://cpc.farnell.com/honest-well/t23a ... dp/SW03177

[oryxhasapi]

or find a pi gpio compatible fan the size of the cpu

[ItsMurf]

Sounds good! Thanks to everyone who responded!

[loverpi]

We came across the overheating issue in our internal engineering review of this product. 207F (97C) with CPU load only in 71F (22C) ambient. We attempted to measure CPU+GPU load but the board crashed on numerous attempts. We saw temperatures go up to 210+F.

Without our heatsinks

img_thermal_1457442329126.jpg (44.98 KiB) Viewed 97939 times

After heatsink treatment: http://amzn.to/1Tqednr, temperature went down to a more manageable 50C.

with our heatsink

img_thermal_1457442285477.jpg (50.56 KiB) Viewed 97939 times

[GTR2Fan]

Nice advert!

Just the one heatsink on the SOC should nearly always be sufficient. Pimoroni are selling a nice one for £1.

[Nickcn]

deleted

[loverpi]

Nice advert!

Just the one heatsink on the SOC should nearly always be sufficient. Pimoroni are selling a nice one for £1.

Our heatsinks are better and cheaper after shipping. You can buy both and test yourself We will include them for free with our Pi 3.

[Pithagoros]

I'm not sure what I'm doing wrong but a fingertip on the SoC occasionally and I'm finding it only slightly warm.

I think it's because I'm just using the thing and not deliberately trying to make it thrash.

[Heater]

You are not doing anything wrong. loverpi is spamming us. See identical posts in other threads.

[loverpi]

You may think it's spam but we consider it a legitimate issue with the board. I'm sure the foundation will put out a new Raspbian image with new firmware to fix the overheating issue by throttling the SoC more aggressively. For the HDMI signal to get corrupted is simply unacceptable. Even if the SoC cools down, the problem won't go away until you reboot.

Run the test yourself: apt-get install -y sysbench && sysbench --num-threads=4 --max-requests=999999 --test=cpu run

Warning: you may damage your board if you have this on for too long. To stop, press control C.

https://www.youtube.com/watch?v=0bC3dQYHfmQ

[jamesh]

You may think it's spam but we consider it a legitimate issue with the board. I'm sure the foundation will put out a new Raspbian image with new firmware to fix the overheating issue by throttling the SoC more aggressively. For the HDMI signal to get corrupted is simply unacceptable. Even if the SoC cools down, the problem won't go away until you reboot.

Run the test yourself: apt-get install -y sysbench && sysbench --num-threads=4 --max-requests=999999 --test=cpu run

Warning: you may damage your board if you have this on for too long. To stop, press control C.

https://www.youtube.com/watch?v=0bC3dQYHfmQ

I suspect your board may have a fault, since you seem to be the only one with any problems as you describe. Statistically, one example is insignificant.

There is no way the chip will fail if it gets over 85, they are tested to 120 and will go higher (they have a reduced lifetime at those temps).

[gkreidl]

You may think it's spam but we consider it a legitimate issue with the board. I'm sure the foundation will put out a new Raspbian image with new firmware to fix the overheating issue by throttling the SoC more aggressively. For the HDMI signal to get corrupted is simply unacceptable. Even if the SoC cools down, the problem won't go away until you reboot.

Run the test yourself: apt-get install -y sysbench && sysbench --num-threads=4 --max-requests=999999 --test=cpu run

Warning: you may damage your board if you have this on for too long. To stop, press control C.

https://www.youtube.com/watch?v=0bC3dQYHfmQ

I've done a one hour job running with 400% CPU, When the temperature reached 80+ C, the RPi started to throttle down and never went beyond 83 C. It was mostly running with a clock speed between 922 and 960 MHz.

[MarkTF]

We came across the overheating issue in our internal engineering review of this product. 207F (97C) with CPU load only in 71F (22C) ambient.

Did you record the SOC's temperature sensor during this test? ( cat /sys/class/thermal/thermal_zone0/temp )

We attempted to measure CPU+GPU load but the board crashed on numerous attempts. We saw temperatures go up to 210+F.

What are you using for a power supply to the board? Inadequate power would also be suspect in this scenario.

After heatsink treatment: temperature went down to a more manageable 50C.

For what it's worth, your picture indicates that the 50 C reading is from the circuit board adjacent to the SOC so the SOC is likely considerably hotter, but is hidden behind the heat sink.

[Fidelius]

I've done a one hour job running with 400% CPU, When the temperature reached 80+ C, the RPi started to throttle down and never went beyond 83 C. It was mostly running with a clock speed between 922 and 960 MHz.

Is this with a Pi3 inside a case or with a "naked" Pi3 ?

I suspect a case, because there I get similar behavior. Without case, under full load the clock changed between 1150 and 1190 MHz in my case (without case) [edit: but with a passive heat-sink on top of the Soc]. But I only used the Linux program "stress --cpu 4" to create a full load on all cores for 15+ minutes. You did some compilation, I suppose?

Some say the produced heat depended on the ARM instructions being use. But isn't an ARM core either being under full 100% load or not? (Another questions is when we started to produce GPU load in parallel to a full ARM load...)

Also I still didn't find a reliable source clarifying the question whether the Broadcom Soc's internal temperature sensor's °C value differs significantly from an external temperature sensor's °C value, as one poster suggested in the other heating thread. Does anybody know?

[Heater]

Fidelius,

Also I still didn't find a reliable source clarifying the question whether the Broadcom Soc's internal temperature sensor's °C value differs significantly from an external temperature sensor's °C value,

Of course they will differ. One is very close to the action, the source of heat, the other is far away.

What on Earth do you expect?

[MarkTF]

[Also I still didn't find a reliable source clarifying the question whether the Broadcom Soc's internal temperature sensor's °C value differs significantly from an external temperature sensor's °C value, as one poster suggested in the other heating thread. Does anybody know?

My understanding is that these sorts of internal temperature sensors aren't particularly accurate. They need to be good enough for the intended purpose of monitoring for potentially damaging conditions, but they aren't calibrated. Somewhere I saw potential errors on the order of 10 C tossed out by an engineer working on a different (not Broadcom) ARM processor.

Keep in mind also that the chip die might not be at a uniform temperature. Generally the logic is segregated into different sections so exercising the CPU(s) with the GPU section essentially idle would produce a temperature gradient across the die and vice versa. If the temperature is sensed on an idle portion of the chip it would be different than the hottest part. I couldn't say how big an effect this would be, but I believe it is significant.

Finally there is the potential for measurement error in the various infrared photos that have been published. We have a tendency to believe numbers on a digital display, but instruments are subject to some level of error, particularly if they are not properly calibrated.

[gkreidl]

I've done a one hour job running with 400% CPU, When the temperature reached 80+ C, the RPi started to throttle down and never went beyond 83 C. It was mostly running with a clock speed between 922 and 960 MHz.

Is this with a Pi3 inside a case or with a "naked" Pi3 ?

I suspect a case, because there I get similar behavior. Without case, under full load the clock changed between 1150 and 1190 MHz in my case (without case). But I only used the Linux program "stress --cpu 4" to create a full load on all cores for 15+ minutes. You did some compilation, I suppose?

Some say the produced heat depended on the ARM instructions being use. But isn't an ARM core either being under full 100% load or not? (Another questions is when we started to produce GPU load in parallel to a full ARM load...)

Also I still didn't find a reliable source clarifying the question whether the Broadcom Soc's internal temperature sensor's °C value differs significantly from an external temperature sensor's °C value, as one poster suggested in the other heating thread. Does anybody know?

1) My Pi 3 is not in a case. And if it would be I would keep the lid open (have done so with all my Pis); it usually makes a difference of about 10 C. No heats sink (I've ordered some).
2) I converted a video with HandBrake and watched from a terminal via SSH (top, measure temp and arm freq. etc).
3) Regarding the GPU: the video decoder doesn't produce much heat, but the 3D part does:
playing a 1080p video with omxplayer: 48 C
playing the same video with kodi, also using omxplayer: 63 C; kodi uses OpenGL ES for the interface.

[java]

What I fail to understand, is here the folks are crying in their milk, that the RPi 3, running stock speeds is overheating. Elsewhere, the folks are doing fairly well overclocking the bejeezers out of their RPi 3's, with and without heatsinks, and mostly smiling.

This does not make much sense to me. I don't have a RPi 3 yet, hopefully the budget will allow me to buy one, come month end.

Got a Pi B, Pi 2 and a Pi Zero - very happy with all of them.

[gkreidl]

What I fail to understand, is here the folks are crying in their milk, that the RPi 3, running stock speeds is overheating. Elsewhere, the folks are doing fairly well overclocking the bejeezers out of their RPi 3's, with and without heatsinks, and mostly smiling.

This does not make much sense to me. I don't have a RPi 3 yet, hopefully the budget will allow me to buy one, come month end.

Got a Pi B, Pi 2 and a Pi Zero - very happy with all of them.

Regarding overclocking the Pi3: it may be possible to overclock the SDRAM or the SD card interface, but overclocking the processor is rather useless, because it will start to throttle down if multiple cores are used to a certain degree.

[Fidelius]

Thank you MarkTF and Gkreidl for your explanations. Very interesting.
(And I forgot to mention next to my throttle-back values, that I have a small passive heat-sink sitting on-top of the SoC. Edited now.)

Also I still didn't find a reliable source clarifying the question whether the Broadcom Soc's internal temperature sensor's °C value differs significantly from an external temperature sensor's °C value,

Of course they will differ. One is very close to the action, the source of heat, the other is far away.

Yes, that's clear, but in the other thread the external and more far away sensor showed 15-20° C more than the internal and closer sensor. :-)
That's what had surprised me, despite the logical explanation that the internal sensor needn't be calibrated an can have errors up to 10° C (like Mark explained).