lordbaronmarshall
Posts: 9
Joined: Wed Apr 06, 2016 5:38 pm

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Sun Apr 10, 2016 12:27 am

I also have one of the RPi3 that heats to 100Celcius. I've found it to be, usually, a problem with Firefox. In particular, watching Youtube with Firefox. I checked the System Monitor and found Youtube and Firefox keep all four CPU at 50%-70%, causing extreme temps within 30 seconds(hot enough to burn skin) YES I HAVE THE HEAT SINKS. The heat is so excessive that it has destroyed data not only on the micro sdcard, but also destroyed data on the USB storage device I was using. I don't think you can justify that at all.

I tried a few browsers and have found that Midori is using the lowest percentages and gathering little to no heat. It still cannot run Youtube videos without the heat problems(although it takes a little longer to heat up with Midori).
If I need to watch Youtube, then I use Midori to get the url, then youtube-dl to download, then omxplayer to watch the video.
I know it sounds like a pain, but this keeps the heat down and CPU percentages around 10%-20%
I've also started trying out streaming URLs with VLC, but don't have any results yet.

gkreidl
Posts: 6355
Joined: Thu Jan 26, 2012 1:07 pm
Location: Germany

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Sun Apr 10, 2016 4:53 am

lordbaronmarshall wrote:I also have one of the RPi3 that heats to 100Celcius. I've found it to be, usually, a problem with Firefox. In particular, watching Youtube with Firefox. I checked the System Monitor and found Youtube and Firefox keep all four CPU at 50%-70%, causing extreme temps within 30 seconds(hot enough to burn skin) YES I HAVE THE HEAT SINKS. The heat is so excessive that it has destroyed data not only on the micro sdcard, but also destroyed data on the USB storage device I was using. I don't think you can justify that at all.

I tried a few browsers and have found that Midori is using the lowest percentages and gathering little to no heat. It still cannot run Youtube videos without the heat problems(although it takes a little longer to heat up with Midori).
If I need to watch Youtube, then I use Midori to get the url, then youtube-dl to download, then omxplayer to watch the video.
I know it sounds like a pain, but this keeps the heat down and CPU percentages around 10%-20%
I've also started trying out streaming URLs with VLC, but don't have any results yet.
If the temperataure in the SOC reaches 80C, the ARM clock is throttled down.
Heat sinks are needed, if you want to get the best speed, but they are quite useless, if there is no airflow, which is the case in most closed plastic boxes.
But that does not destroy data on the SD card or even on a USB stick. This must be caused by something else.
Minimal Kiosk Browser (kweb)
Slim, fast webkit browser with support for audio+video+playlists+youtube+pdf+download
Optional fullscreen kiosk mode and command interface for embedded applications
Includes omxplayerGUI, an X front end for omxplayer

ejolson
Posts: 6044
Joined: Tue Mar 18, 2014 11:47 am

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Sun Apr 10, 2016 4:46 pm

gkreidl wrote:Heat sinks are needed, if you want to get the best speed, but they are quite useless, if there is no airflow, which is the case in most closed plastic boxes.
But that does not destroy data on the SD card or even on a USB stick. This must be caused by something else.
I think the implication is that data is lost by the sdcard melting and/or too much heat physically damaging it.

gkreidl
Posts: 6355
Joined: Thu Jan 26, 2012 1:07 pm
Location: Germany

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Sun Apr 10, 2016 9:11 pm

ejolson wrote:
gkreidl wrote:Heat sinks are needed, if you want to get the best speed, but they are quite useless, if there is no airflow, which is the case in most closed plastic boxes.
But that does not destroy data on the SD card or even on a USB stick. This must be caused by something else.
I think the implication is that data is lost by the sdcard melting and/or too much heat physically damaging it.
Wrong implication, and wrong fact: Neither the SD card nor the UDB stick were reported to melt. Data corruption can be cause by a lot of things.
Minimal Kiosk Browser (kweb)
Slim, fast webkit browser with support for audio+video+playlists+youtube+pdf+download
Optional fullscreen kiosk mode and command interface for embedded applications
Includes omxplayerGUI, an X front end for omxplayer

User avatar
Rive
Posts: 586
Joined: Sat Mar 26, 2016 5:21 pm
Location: USA

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Sun Apr 10, 2016 10:28 pm

gkreidl wrote:
ejolson wrote:
gkreidl wrote:Heat sinks are needed, if you want to get the best speed, but they are quite useless, if there is no airflow, which is the case in most closed plastic boxes.
But that does not destroy data on the SD card or even on a USB stick. This must be caused by something else.
I think the implication is that data is lost by the sdcard melting and/or too much heat physically damaging it.
Wrong implication, and wrong fact: Neither the SD card nor the UDB stick were reported to melt. Data corruption can be cause by a lot of things.

How about this scenerio?


Pi overheats, freezes up/locks up and is then manually powered off. then on. Could that cause a dirty bit and/or flipped bit and subsequent sdcard data corruption?
DNPNWO

ejolson
Posts: 6044
Joined: Tue Mar 18, 2014 11:47 am

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Sun Apr 10, 2016 10:39 pm

gkreidl wrote:
ejolson wrote:
gkreidl wrote:Heat sinks are needed, if you want to get the best speed, but they are quite useless, if there is no airflow, which is the case in most closed plastic boxes.
But that does not destroy data on the SD card or even on a USB stick. This must be caused by something else.
I think the implication is that data is lost by the sdcard melting and/or too much heat physically damaging it.
Wrong implication, and wrong fact: Neither the SD card nor the UDB stick were reported to melt. Data corruption can be cause by a lot of things.
Oops, the melting sdcard must have been in a different thread. As the Linux kernel is pretty robust, these days most data corruption is caused by some sort of hardware malfunction. There is a thread on incorrect results which seems to indicate that many Pi 3 computers experience random hardware faults that yield incorrect answers when solving systems of linear equations using code that has been optimized with the ARM NEON instructions. Underclocking, heat sinks and overvolting seem to help. If I were experiencing any kind of data corruption on a Pi 3, I'd first check to make sure the linpack test binary gave correct results before looking for other problems.

User avatar
davidcoton
Posts: 5513
Joined: Mon Sep 01, 2014 2:37 pm
Location: Cambridge, UK
Contact: Website

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Mon Apr 11, 2016 8:44 am

ejolson wrote: There is a thread on incorrect results which seems to indicate that many Pi 3 computers experience random hardware faults that yield incorrect answers when solving systems of linear equations using code that has been optimized with the ARM NEON instructions. Underclocking, heat sinks and overvolting seem to help. If I were experiencing any kind of data corruption on a Pi 3, I'd first check to make sure the linpack test binary gave correct results before looking for other problems.
Let's ignore your personal gripes for the moment -- I'm not dismissing the errors you found, but they are not primary here.

@lordbaronmarshall
The first question is, when the Pi should protect itself from overheating, why does it get so hot (100C)? When did you observe that? How did you check temp?
Things to check:
What power supply and cable is in use?
What peripherals are connected?
Is it place uncased on a conductive surface (metal)?

If that is all OK, I suspect a faulty Pi. But it is becoming clear that there is a big range of thermal behaviour on different Pi3s. Some proper investigation of a substantial sample is necessary, and some way to identify affected models by serial number or other observable facts. If there is a manufacturing problem, I hope RPF/Sony/RS/Farnell will be able to clarify things soon.

The second question is, how does the data on sdcard AND USB storage get damaged?
That almost certainly due to power problems. If the Pi gets that hot, it is drawing excess power. Which probably means reduced voltage, which is known to affect sdcard and USB devices.
Are there any low voltage warnings (power LED off, rainbow square top right of screen)?
Subsidiary -- is the software up to date? (I think the first RPi3 Raspian didn't monitor voltage properly.)
Location: 345th cell on the right of the 210th row of L2 cache

loadbang
Posts: 36
Joined: Mon Aug 13, 2012 4:56 pm

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Mon Apr 18, 2016 9:56 am

After running my Pi 3 under load using 'stress -i 4 -c4' and finding it gets to 80ºC not overclocked and freezing requiring a reboot, I went out and got a set of these:

Image

1.35GHz is stable at 78ºC. Only using the larger heatsink, the other two not required on the 3 unlike the first generation Pi's.

I have just salvaged this from a Pentium 1 motherboard. I need some thermal tape before I can test:

Image

Image

Image

ejolson
Posts: 6044
Joined: Tue Mar 18, 2014 11:47 am

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Tue Apr 19, 2016 7:16 pm

davidcoton wrote:
ejolson wrote:If I were experiencing any kind of data corruption on a Pi 3, I'd first check to make sure the linpack test binary gave correct results before looking for other problems.
The second question is, how does the data on sdcard AND USB storage get damaged?
If the CPU experiences computational errors randomly while it is running, this can easily lead to corrupted data being written to the sdcard. Since complete system lock up has been observed by running computationally intensive but otherwise bug free software on the Pi 3, it is not surprising that data corruption on an sdcard is also observed. System stress tests that check correctness of computational results are important because they can help diagnose the cause of random data corruption that occurs at a lower rate of incidence under other operating conditions. In particular, if the hardware is so broken it can't solve systems of linear equations using standard subroutine libraries, then there is little reason to be surprised when data corruption occurs while performing other tasks. The same hardware is being used in both cases, the only difference is that it is easier to detect when data corruption occurs while solving linear equations because the task is simpler and the result easier to check.

As many Pi 3 computers experience random data corruption under stock voltage and frequency settings when solving systems of linear equations, this program seems ideally suited for trouble shooting data corruption issues which may appear when performing other tasks.

User avatar
Rive
Posts: 586
Joined: Sat Mar 26, 2016 5:21 pm
Location: USA

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Tue Apr 19, 2016 7:30 pm

loadbang wrote:After running my Pi 3 under load using 'stress -i 4 -c4' and finding it gets to 80ºC not overclocked and freezing requiring a reboot...
1.35GHz is stable at 78ºC. Only using the larger heatsink, the other two not required on the 3 unlike the first generation Pi's.

I have just salvaged this from a Pentium 1 motherboard. I need some thermal tape before I can test:

Image
I have resolved my overheating issue. At first i had one of those small little copper heatsinks...might shave off 5C or so (if you are lucky), but if you have nothing to move the heat off it, it is useless (like if in a case).

BEFORE:
Image

Stress isnt really anything (anymore for me). I also replaced my heatsink. I can do Stress at 43C/44C. The trick is now, how well can your pi3 handle something like Cpuburn-a53 which uses NEON and is designed to heat it up? That is the real trick to see if your pi3 cooling is adequate or not. If yours is not adequate, you may need a fan.

I can run Cpuburn-a53 and stay at around 63C/64C with no throttling (and i am overclocked at 1.3 GHz).

See: viewtopic.php?p=952371#p952371

AFTER:
Image

Image

Image
Last edited by Rive on Tue Apr 19, 2016 8:32 pm, edited 3 times in total.
DNPNWO

ejolson
Posts: 6044
Joined: Tue Mar 18, 2014 11:47 am

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Tue Apr 19, 2016 7:49 pm

Rive wrote:
loadbang wrote:After running my Pi 3 under load using 'stress -i 4 -c4' and finding it gets to 80ºC not overclocked and freezing requiring a reboot...
1.35GHz is stable at 78ºC. Only using the larger heatsink, the other two not required on the 3 unlike the first generation Pi's.

I have just salvaged this from a Pentium 1 motherboard. I need some thermal tape before I can test:
Stress isnt really anything. I also replaced my heatsink. I can do Stress at 43C/44C. The trick is now, how well can your pi3 handle something like Cpuburn-a53 which uses NEON and is designed to heat it up? That is the real trick to see if your pi3 cooling is adequate or not. If yours is not adequate, you may need a fan.

I can run Cpuburn-a53 and stay at around 63C/64C with no throttling (and i am overclocked at 1.3 GHz).
Since cpuburn-a53 does not check whether data corruption or computational errors occur during the run, it is more useful as a space heater to determine maximum CPU temperatures under load then for diagnosing data corruption issues that might result from improper cooling or over clocking. It's still a useful program, though.

timrowledge
Posts: 1354
Joined: Mon Oct 29, 2012 8:12 pm
Location: Vancouver Island
Contact: Website

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Tue Apr 19, 2016 11:35 pm

So do y'all reckon this is a good enough heatsink for my Pi3?
IMG_0259.jpg
IMG_0259.jpg (55.19 KiB) Viewed 8292 times
Making Smalltalk on ARM since 1986; making your Scratch better since 2012

mi7chy
Posts: 27
Joined: Mon Mar 28, 2016 5:50 am

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Wed Apr 20, 2016 6:52 am

Picked up a 2nd RPi3 that draws more power causing the rainbow square to flash, idles at ~62C with default 'on demand' setting and shoots up to 83C just playing a little windowed 360p video. Huge discrepancy compared to the 1st RPi3 with identical attachments. This isn't just hot but defective.

User avatar
RaTTuS
Posts: 10616
Joined: Tue Nov 29, 2011 11:12 am
Location: North West UK
Contact: Twitter YouTube

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Wed Apr 20, 2016 8:25 am

mi7chy wrote:Picked up a 2nd RPi3 that draws more power causing the rainbow square to flash, idles at ~62C with default 'on demand' setting and shoots up to 83C just playing a little windowed 360p video. Huge discrepancy compared to the 1st RPi3 with identical attachments. This isn't just hot but defective.
so RMA it
How To ask Questions :- http://www.catb.org/esr/faqs/smart-questions.html
WARNING - some parts of this post may be erroneous YMMV

1QC43qbL5FySu2Pi51vGqKqxy3UiJgukSX
Covfefe

User avatar
Gerd
Posts: 66
Joined: Wed Mar 16, 2016 10:48 am
Location: Europe

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Wed Apr 20, 2016 9:13 am

loadbang wrote: I have just salvaged this from a Pentium 1 motherboard. I need some thermal tape before I can test:
Image
Before applying thermal tape, you need to cut one edge from the cooler, because the voltage smds are higher than the cpu, so the cooler is not lying planar on the cpu.
But go for it! It is all about having a large surface.

Sneak preview of my installation:
Attachments
Raspi Sneak Preview.jpg
Raspi Sneak Preview.jpg (48.67 KiB) Viewed 8138 times

gregeric
Posts: 1509
Joined: Mon Nov 28, 2011 10:08 am

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Wed Apr 20, 2016 6:42 pm

I'm also salvaging old PC parts for cooling. But the aluminium heat sink itself was too tall, so...
HeatSink.jpg
HeatSink.jpg (57.51 KiB) Viewed 8033 times
A 1993 vintage, ceramic packaged TI 486 :)
Fitted.jpg
Fitted.jpg (63.5 KiB) Viewed 8033 times
Perfectly sized to fit, but a metal shim is needed to lift it higher than the SMPS components.
Profile.jpg
Profile.jpg (63.03 KiB) Viewed 8033 times
Just enough clearance for a HAT too.

Edited to add:

Initial sysbench stress tests, uncased, orientation horizontal, propped up to give airflow underneath, & ambient 25C: no throttling, temp rising eventually to 75C. That's with a Unicorn HAT mounted & running the clock demo, a HAT which itself generates a bit of heat. Temperature dropped ~6C orienting it vertically so air could convect easily between the two boards.

ejolson
Posts: 6044
Joined: Tue Mar 18, 2014 11:47 am

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Fri Apr 22, 2016 7:01 pm

mi7chy wrote:Picked up a 2nd RPi3 that draws more power causing the rainbow square to flash, idles at ~62C with default 'on demand' setting and shoots up to 83C just playing a little windowed 360p video. Huge discrepancy compared to the 1st RPi3 with identical attachments. This isn't just hot but defective.
Did you RMA the defective Pi 3 and was the replacement any better?

The fact that the performance of the Pi 3 strongly depends on heat sink, voltage and silicon variations means each student will get different timing results when running the same algorithm using the same data on different Pi 3 computers. This could make it difficult to use the Pi 3 when teaching efficiency of algorithms and numeric methods.

Pithagoros
Posts: 578
Joined: Wed Nov 12, 2014 8:16 pm

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Fri Apr 22, 2016 7:31 pm

timrowledge wrote:So do y'all reckon this is a good enough heatsink for my Pi3?
IMG_0259.jpg
Liquid is where it is at:

Image

blc
Posts: 465
Joined: Mon Sep 05, 2011 9:28 am

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Sat Apr 23, 2016 12:52 am

Pithagoros wrote:
timrowledge wrote:So do y'all reckon this is a good enough heatsink for my Pi3?
IMG_0259.jpg
Liquid is where it is at:

Image
You may be in jest, but that's an impressive cooling rig; a lot of work has gone into that.

mi7chy
Posts: 27
Joined: Mon Mar 28, 2016 5:50 am

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Sat Apr 23, 2016 7:32 am

ejolson wrote:Did you RMA the defective Pi 3 and was the replacement any better?

The fact that the performance of the Pi 3 strongly depends on heat sink, voltage and silicon variations means each student will get different timing results when running the same algorithm using the same data on different Pi 3 computers. This could make it difficult to use the Pi 3 when teaching efficiency of algorithms and numeric methods.
All valid concerns. Another thing to consider is due to variances in power consumption not all RPi3s will run on existing 5V 2A power supplies so in a classroom environment you'll be looking at shelling out bucks for new 2.5A power supplies. My first RPi3 ran fine on 5V 2A while the 2nd flashes rainbow square during initial setup.

As for RMA, the place I bought it from is out of stock but crossing my fingers the new stock will be better. Worst case I'll just wait for RPi4 since there's no thermal headroom with 40nm. It should be at least 32nm with passive heat sink if not 28nm like the Odroid C2 which runs at about 62C under the same conditions that causes 82C on RPi3.
Last edited by mi7chy on Sat Apr 23, 2016 3:50 pm, edited 1 time in total.

marked
Posts: 218
Joined: Fri Jul 29, 2011 4:25 pm

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Sat Apr 23, 2016 8:01 am

Pithagoros wrote:Image
Looks like someone has been playing around in the Umbrella Labs, and is about to infect themselves with the T-Virus.

edit. And following through to the original page, I find I am not the only one to have thought that.

timrowledge
Posts: 1354
Joined: Mon Oct 29, 2012 8:12 pm
Location: Vancouver Island
Contact: Website

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Sat Apr 23, 2016 3:39 pm

Pithagoros wrote: Liquid is where it is at:
You win!

Now, replace the PiB with a 3, the plastic piping with proper soldered copper, and add a steam engine drive circulation pump/fan and we're done :D
Making Smalltalk on ARM since 1986; making your Scratch better since 2012

mi7chy
Posts: 27
Joined: Mon Mar 28, 2016 5:50 am

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Sat Apr 23, 2016 10:19 pm

ejolson wrote:Did you RMA the defective Pi 3 and was the replacement any better?
My local store got new stock in so I went ahead and purchased a 3rd (UK made RS/Allied Electronics compared to PRC for Element14) to support the RPi foundation. Now on to the quick analysis. Not a reliable indicator of when it was manufactured but it has a date code of 1603 (3rd week of 2016) on the ethernet housing compared to 1604 for the 1st good and 1553 for 2nd furnace. The date code isn't newer than the 1st I bought about a month ago so they must manufacture in bulk and have sitting in a warehouse. So far the 3rd behaves acceptably like the 1st. Playing the same 360p video in a window it doesn't hit 82C within the 1st minute and doesn't throttle down to 900MHz range like the 3rd. Temperature does slowly rise, though, to a point in the second loop of the 3m51s video where it starts throttling between 1000 and 1100MHz range. This I think I can live with since adding a passive heat sink should help extend the time before thermal throttling occurs for most work loads in an open case.

loadbang
Posts: 36
Joined: Mon Aug 13, 2012 4:56 pm

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Wed Apr 27, 2016 9:14 am

Gerd wrote:
loadbang wrote: I have just salvaged this from a Pentium 1 motherboard. I need some thermal tape before I can test:
Image
Before applying thermal tape, you need to cut one edge from the cooler, because the voltage smds are higher than the cpu, so the cooler is not lying planar on the cpu.
But go for it! It is all about having a large surface.

Sneak preview of my installation:
I noticed this. Placed some kapton tape on the bottom and then offset the heatsink slightly to avoid them.

I still haven't ordered the tape. Waiting to need to buy something else form CPC to avoid carriage costs. Trying to find a decent fan that the RPi can also power.

loadbang
Posts: 36
Joined: Mon Aug 13, 2012 4:56 pm

Re: RPi 3 SoC at nearly 100 °C, reporting 80 °C

Wed Apr 27, 2016 9:29 am

For all those running stress tests and measuring CPU speed, make sure you are running to get your speed:

# vcgencmd measure_clock arm | sed -e "s/frequency(45)=//g"

rather than;

# cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_cur_freq

First one shows the actual CPU frequency, the other is what the governor sets.

Return to “General discussion”