Page 2 of 10

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

Posted: Wed Mar 02, 2016 12:59 pm
by allfox
I made a bad joke. Deleted.

Thanks to moderators to make those information clear.

So Pi 3 is hot, but that's all of it. It's tested for that.
Heat sink would help to keep Pi 3 at full speed. But without it, Pi 3 still is solid, just might be slower at sometime.
It won't fire, won't blow up.

I'm thinking that maybe some plastic case could melt. And I won't touch the chip.

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

Posted: Wed Mar 02, 2016 1:04 pm
by Fidelius
jamesh wrote:Ray tracing is very compute intensive, so I would suggest a heatsink. Use the one you have, it should be fine. The tape *should* transfer the heat.
Thanks for this info, it sounds good. Will do so, in order to keep the ARM8 running at its 1200 MHz also during compute intensive tasks. :-)

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

Posted: Wed Mar 02, 2016 1:17 pm
by Fidelius
Concerning the mentioned difference of 17°/20 °C between the Videocore's temperature sensor near the ARM8, and external heat cameras:

Does the Pi3's internal temperature sensor always report a value being 17°/20°C below external sensors? I.e. is it linear, or "only" when the ARM is getting hotter?

Because in desktop idle mode my Pi3 shows around ~45° C with "vcgencmd measure_temp". It's a similar value compared to some AMD/Intel desktop PCs I use.

Now I know that we can't directly compare ARM and AMD/Intel CPUs. However, if we had to add +17°/+20°C to this 45°C value in desktop idle mode, wouldn't a Pi3 running at 62°/65° C in desktop idle mode be relatively hot for not-so-much-doing? :-)

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

Posted: Wed Mar 02, 2016 1:25 pm
by Fidelius
rurwin wrote:The Pi will start to throttle-back when the temperature sensor reaches 85C, so you will want to avoid getting too close to that figure simply to maintain performance. However there is not going to be any significant danger to the SoC running at that sort of temperature or even consideraby hotter than that. Testing has been done at well over 100C without obvious damage although that may deminish the longevity of the chip and it could give you a nasty burn if you were to touch it.

I wouldn't worry about the placement of the sensor making the CPU temperature inaccurate. Those 20 degrees C are not going to damage anything.
These are informations good to know, thanks!

My Pi3 starts at ~80° C to throttle-back, which we can see nicely with a two open terminals, one running:

Code: Select all

while true; do vcgencmd measure_temp; sleep 1; done
... and another running:

Code: Select all

while true; do vcgencmd measure_clock  arm; sleep 1; done
Then we start a stress test, like for example:

Code: Select all

stress --cpu 4 --timeout 600
When the ARM reaches 80°, its clock decreases from 1200 MHz towards 1000 MHz slowly and nicely, and when the temperature starts to fall then the clock goes up slowly again, and after a while the clock and temperature level out at around 80° with a clock varying from 1000 to 1200 MHz. Very interesting!

(P.S. And the Broadcom chip sensor's 80° making the ARM to throttle-back, is 97°/100° C from external view?)

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

Posted: Wed Mar 02, 2016 2:36 pm
by chrisoh
rurwin wrote:
chrisoh wrote:There seems to be an inconsistancy somewhere. Gareth Halfacree's findings indicate the Pi's internal sensor reporting < 80C when it's actually at almost 100C, well past the auto-throttling stage and into the 'will possibly damage the chip' range. When the Pi thinks it has reached 85C, it's actually already at 110C or possible 120C+.
Read it again. He found a temperature differential of 17C. If the sensor is reading 85C then the camera may see 102C.
Where have you read that 100C will possibly damage the chip?
No need, but thanks.
Gareth has indicated that the temperature discrepancy increases with an increase in chip temp. Assuming the internal sensor isn't completely dud and reads 38C when it is actually 38C, it's fairly straightforward to extrapolate the temperatures observed by Gareth. My back-of-a-napkin calulations shows that the auto-throttling will kick in when the observed (not internally sensed) temperature is from anywhere between 99.55C and 111.65C.
I give a range because I've seen 3 different statements about the actual auto-throttling temperature (80C (Fidelius), 82C (Gareth) and 85C (yourself)).
I think the core of all this is that Gareth measured ~100C on the chip and the auto-throttling hadn't started.

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

Posted: Wed Mar 02, 2016 4:10 pm
by ric96
jamesh wrote:
I'm not using one, seems fine to me. Why increase the cost for everyone by including it?
Assuming you live in the UK, the climate their is much cooler. During peak summers ind India, my pi 2 easily hits ~70c at full load and around ~60c during winters. Perfectly Good chance that the 3 will thermal throttle here during summers.
Although selling their own might be a good money spinner. Make it in the shape of a Raspberry or something like that (patent applied for!)
Sweet. But in the mean time are the generic ones any good? http://www.amazon.in/dp/B0110J9QFY/ref= ... uctDetails

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

Posted: Wed Mar 02, 2016 4:35 pm
by Fidelius
ric96 wrote:But in the mean time are the generic ones any good? www.amazon.in
Your linked ARM heatsink with a Pi case looks very similar to the heatsink which came with my Pi case (please see my previous post linking to it here in the same thread).

In any case James' advice worked for me:
jamesh wrote:Ray tracing is very compute intensive, so I would suggest a heatsink. Use the one you have, it should be fine. The tape *should* transfer the heat.
With the previously explained

Code: Select all

stress --cpu 4
... on a Pi3 with a case but without heatsink, the internal 80° C sensor value triggering the ARM's throttle-back was reached within one or two minutes and leveled out at the ARM throttled to ~1000 MHz.

Now still with a case on but containing the linked heat-sink, the internal 80° C sensor value is being reached much later (~15 minutes) and levels out at the ARM throttled-back to ~1050 MHz.

When I open the top of the case, exposing the heatsink, the Pi3 still goes to 80° C but only a little throttling is going on: it levels out at around ~1190 MHz.

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

Posted: Wed Mar 02, 2016 4:38 pm
by jamesh
ric96 wrote:
jamesh wrote:
I'm not using one, seems fine to me. Why increase the cost for everyone by including it?
Assuming you live in the UK, the climate their is much cooler. During peak summers ind India, my pi 2 easily hits ~70c at full load and around ~60c during winters. Perfectly Good chance that the 3 will thermal throttle here during summers.
Although selling their own might be a good money spinner. Make it in the shape of a Raspberry or something like that (patent applied for!)
Sweet. But in the mean time are the generic ones any good? http://www.amazon.in/dp/B0110J9QFY/ref= ... uctDetails
Any old heatsink would do. I've used one from a old PC motherboard that was a bit big, but worked. (As a joke btw as was on an older device that didn't need the heatsink)

Image

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

Posted: Wed Mar 02, 2016 4:43 pm
by Pithagoros
Hmm, well after playing with the 3 for a day or so I'm happy with the bluetooth and heatsink and the general reduction in cable spaghetti, but I'm not so impressed with the increase in heat and associated power waste.

So my question is, can I UNDERclock the Pi3, maybe take it down to 900 or 1000, is there a downside to this apart from a slight performance drop?

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

Posted: Wed Mar 02, 2016 4:47 pm
by ric96
Fidelius wrote:
ric96 wrote:But in the mean time are the generic ones any good? www.amazon.in
Your linked ARM heatsink with a Pi case looks very similar to the heatsink which came with my Pi case (please see my previous post linking to it here in the same thread).

In any case James' advice worked for me:
jamesh wrote:Ray tracing is very compute intensive, so I would suggest a heatsink. Use the one you have, it should be fine. The tape *should* transfer the heat.
With the previously explained

Code: Select all

stress --cpu 4
... on a Pi3 with a case but without heatsink, the internal 80° C sensor value triggering the ARM's throttle-back was reached within one or two minutes and leveled out at the ARM throttled to ~1000 MHz.

Now still with a case on but containing the linked heat-sink, the internal 80° C sensor value is being reached much later (~15 minutes) and levels out at the ARM throttled-back to ~1050 MHz.

When I open the top of the case, exposing the heatsink, the Pi3 still goes to 80° C but only a little throttling is going on: it levels out at around ~1190 MHz.
Can you try with a tiny 5v fan if possible, they usually draw ~0.1A so just plug that into the 5v pin.
The work well with my rose apple pi (raspi ripoff that heats a lot) very similar to what i am seeing here, the heasink only helps in delaying the heatup but since their is no airflow it also eventually reaches the limit. With a tiny fan directly over the heatsink, the temps won't go over ~60c worst case.

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

Posted: Wed Mar 02, 2016 4:52 pm
by GTR2Fan
This 14x14x10mm heatsink is about as big as you can go without overhanging the SOC...

http://www.ebay.co.uk/itm/121655033206? ... EBIDX%3AIT

This one knocked around 10°C off my overclocked Pi2B when it was within a gnat's of throttling. Mine's enclosed in a ventilated case, but I'd imagine it would probably be a lot more effective in free air.

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

Posted: Wed Mar 02, 2016 8:53 pm
by ssvb
Has anyone actually tested the Raspberry Pi 3 throttling behaviour with heavier workloads (such as some decently NEON optimized software or a cpuburn application)? I have already asked this question in viewtopic.php?p=916941#p916941

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

Posted: Wed Mar 02, 2016 9:53 pm
by solar3000
jamesh wrote:
ric96 wrote:
jamesh wrote:
I'm not using one, seems fine to me. Why increase the cost for everyone by including it?
Assuming you live in the UK, the climate their is much cooler. During peak summers ind India, my pi 2 easily hits ~70c at full load and around ~60c during winters. Perfectly Good chance that the 3 will thermal throttle here during summers.
Although selling their own might be a good money spinner. Make it in the shape of a Raspberry or something like that (patent applied for!)
Sweet. But in the mean time are the generic ones any good? http://www.amazon.in/dp/B0110J9QFY/ref= ... uctDetails
Any old heatsink would do. I've used one from a old PC motherboard that was a bit big, but worked. (As a joke btw as was on an older device that didn't need the heatsink)

Image
waaa! what is that monster?!
I don't have the pi3 yet and I hope I can get by with a smaller heat sink.
I wonder if I can use a copper/aluminum block to make contact with the CPU and perhaps an aluminum case or cover....

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

Posted: Wed Mar 02, 2016 11:36 pm
by neomew
Hello everyone :)
I'm still a bit new to raspberry pi's and just ordered a pi 3 on Monday.

What kind of heatsink or case or fan should I get for my Raspberry Pi?

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

Posted: Thu Mar 03, 2016 12:14 am
by rurwin
First, don't worry about it. You have to be running some particularly tough software to notice any issues.

Second, get the sort of case that's a square box rather than one moulded around the Pi, with plenty of ventilation.

Third, if you decide you do want a heatsink, a Google search on "Rapberry Pi heatsink" will find plenty of them.

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

Posted: Thu Mar 03, 2016 9:09 am
by jamesh
Surprised no-one has commented on the Pi underneath that huge heatsink.

Check out the SD card holder...that was hand soldered on by Eben I think....which makes it one of the first 10.....

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

Posted: Thu Mar 03, 2016 9:21 am
by Heater
Is there anyway to fix the problem that when people post 3840 x 2160 resolution images here it's un-viewable in the Chrome browser?

Not to mention that it's nearly a one megabyte download which is painful on mobile devices.

And then someone quotes the post containing the image so the layout gets mangled twice!

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

Posted: Thu Mar 03, 2016 9:28 am
by jamesh
Heater wrote:Is there anyway to fix the problem that when people post 3840 x 2160 resolution images here it's un-viewable in the Chrome browser?

Not to mention that it's nearly a one megabyte download which is painful on mobile devices.

And then someone quotes the post containing the image so the layout gets mangled twice!
Dunno, not sure why the image came out so large, when if you click on it, it's smaller. If I have time I'll try and shrink it.

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

Posted: Thu Mar 03, 2016 10:33 am
by jahboater
I like this one for £1, easy to peel off the thermal backing tape, sticks firmly, and appears very effective.
http://thepihut.com/collections/raspber ... i-heatsink
The best position, if you have a choice, is vertical/vertical, that is, the board held upright and the fins aligned to allow the free flow of air from convection. If its not in a case, you could stick one on the sdram too.

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

Posted: Thu Mar 03, 2016 10:39 am
by GTR2Fan
jahboater wrote:I like this one for £1, easy to peel off the thermal backing tape, sticks firmly, and appears very effective.
http://thepihut.com/collections/raspber ... i-heatsink
Nice. That's very similar to the one on my Pi2B that drops load temps by around 10°C in a ventilated case. Have you done any measurements with and without in free air?

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

Posted: Thu Mar 03, 2016 10:49 am
by jahboater
GTR2Fan wrote:
jahboater wrote:I like this one for £1, easy to peel off the thermal backing tape, sticks firmly, and appears very effective.
http://thepihut.com/collections/raspber ... i-heatsink
Nice. That's very similar to the one on my Pi2B that drops load temps by around 10°C in a ventilated case. Have you done any measurements with and without in free air?
I did some time ago, and the temps during stress testing an overclock dropped a few degrees, I cant remember how much exactly but the extra stability enabled me to drop the overvoltage from 6 to 2 which gave a further few degrees drop in temperature, so it now runs in the upper fifties when heavily loaded compared to the upper sixties (where it needed ov6 to avoid crashing). Overall, yes, probably around 10C.
The delta-T at idle is now 8C!.

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

Posted: Thu Mar 03, 2016 11:24 am
by Fidelius
In the other similar thread (Pi 3 Overheating Issue?) which now links to here, there's an interesting pointer to an article named The Raspberry Pi 3 does not halt and catch fire, stating :
Testing two Raspberry Pi 3 boards — the first I received from the Foundation in the run up to Monday’s launch and the second a retail board I picked up through normal channels — I went ahead and tried to replicate Gareth’s results.

Following his methodology I used a thermal camera, although as my thermal camera really isn’t as good as his thermal camera, I also checked my temperature measurements both with an IR thermometer, and a contact probe thermometer.

I couldn’t replicate the result, instead I found a maximum operating temperature of just over 62°C for the BCM2837 at full and sustained load. I was also saw a direct correlation between the temperature reported by the Raspberry Pi 3’s temperature sensor and the temperature reported by my IR thermometer aimed at the SoC — at least to within a couple of degrees.

Crossing my fingers I briefly poked the chip. It’s certainly not 100°C, and after sustained contact my readings of around 60°C seem reasonable. So that’d be “ouch” rather than “argh.” The obvious guess is one of us has either been lucky, or unlucky, in the silicon lottery with the boards we’d received. Because while sensors can be miscalibrated, the finger test is hard to fool.
Gareth Halfacree wrote:I’m in touch with the Foundation, and they’ve been running some tests. Using the same benchmark, their test Pi 3 has hit 83°C external temperature as measured by a contact probe in the upper corner of the SoC – lower than mine but higher than yours — but started to correctly throttle itself and bounce between 77°C and 83°C until the benchmark was killed. So, we’ve got my Pi 3s which hit 100°C and don’t throttle; your Pi 3s which hit 60°C and don’t need to throttle; and the test Pi 3 at the Foundation, which hits 83°C and correctly throttles.

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

Posted: Thu Mar 03, 2016 11:42 am
by GTR2Fan
Or to highlight the part that really matters...
So, we’ve got my Pi 3s which hit 100°C and don’t throttle; your Pi 3s which hit 60°C and don’t need to throttle; and the test Pi 3 at the Foundation, which hits 83°C and correctly throttles.
It's the apparent extremely wide spread in reported temperatures when accurately measured under similar workloads that has people a little uncertain at the moment. As all of these results appear to be genuine and valid, nobody has the faintest idea where they stand with regards to whether they really need a heatsink or not.

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

Posted: Thu Mar 03, 2016 11:52 am
by RaTTuS
I've done some tests on mine :-
sudo apt-get update && sudo apt-get install sysbench -y && sudo apt-get upgrade -y
then

while true; do vcgencmd measure_clock arm; vcgencmd measure_temp; sleep 1; done& sysbench --num-threads=8 --test=cpu --cpu-max-prime=10000000000 run

Code: Select all

[1] 4728
frequency(45)=600000000
temp=43.3'C
sysbench 0.4.12:  multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 8

Doing CPU performance benchmark

Threads started!
frequency(45)=600000000
temp=44.4'C
frequency(45)=600000000
temp=46.5'C
frequency(45)=600000000
temp=47.1'C
frequency(45)=600000000
temp=47.6'C
frequency(45)=600000000
temp=48.2'C
frequency(45)=600000000
temp=48.7'C
frequency(45)=600000000
temp=49.2'C
frequency(45)=600000000
temp=49.2'C
frequency(45)=600000000
temp=49.8'C
frequency(45)=600000000
temp=50.3'C
frequency(45)=600000000
temp=50.3'C
frequency(45)=600000000
temp=50.3'C
frequency(45)=600000000
temp=50.8'C
frequency(45)=600000000
temp=50.8'C
frequency(45)=600000000
temp=50.8'C
frequency(45)=600000000
temp=51.9'C
frequency(45)=600000000
temp=51.4'C
frequency(45)=600000000
temp=51.9'C
frequency(45)=600000000
temp=51.9'C
frequency(45)=600000000
temp=51.4'C
frequency(45)=600000000
temp=51.9'C
frequency(45)=600000000
temp=51.9'C
frequency(45)=600000000
temp=51.9'C
frequency(45)=600000000
temp=51.9'C
frequency(45)=1200000000
temp=53.0'C
frequency(45)=600000000
temp=52.5'C
frequency(45)=600000000
temp=51.9'C
frequency(45)=600000000
temp=52.5'C
frequency(45)=600000000
temp=52.5'C
frequency(45)=600000000
temp=53.0'C
followed by ctrl-c and
kill %1
to kill the jobs [YMMV]

so I have something funky going on on my Pi3

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

Posted: Thu Mar 03, 2016 11:54 am
by gordon77
I would say if you think you need one then fit one.

I don't think I need one.