Page 5 of 10

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

Posted: Fri Mar 04, 2016 2:13 pm
by jdb
RaTTuS wrote:with a 2.5A PSU and good quality lead ,
I boot to command line
open a new terminal and type
watch '(vcgencmd measure_temp; vcgencmd measure_clock arm)'
I get that oscillating between 1.2G and .6G [1200000000 and 600000000 ]
temp is 45.5'C approx
with nothing else running
the power light does not flash :-(

dang I'm going to have to fiddle more :(
If your Pi CPUs are doing nothing, the default clock is 600MHz. The ondemand governor only requests 1.2GHz when >=1 core is busy.

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

Posted: Fri Mar 04, 2016 2:46 pm
by RaTTuS
jdb wrote: If your Pi CPUs are doing nothing, the default clock is 600MHz. The ondemand governor only requests 1.2GHz when >=1 core is busy.
doh - of cause..
ok when it is measuring 82 / 83 deg C it starts scaling back from 1.2G downwards then the temp drops and it scales upwards

the external reading from a IR thermometer is 86.7 deg C max - I know this is not the best measurement but that seems OKish to me [careful placing of hands , small movements to get the highest figure etc...]

[edit for clarity]
^ this is what happens now when I run
sysbench --num-threads=8 --test=cpu --cpu-max-prime=10000000000 run
on another terminal
so I don't get the heat as described in the OP post and https://twitter.com/ghalfacree but I don't have the required equipment to measure external temps accurately enough
but I do see the correct stuff happening in the CPU

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

Posted: Fri Mar 04, 2016 3:23 pm
by jamesh
RaTTuS wrote:with a 2.5A PSU and good quality lead ,
I boot to command line
open a new terminal and type
watch '(vcgencmd measure_temp; vcgencmd measure_clock arm)'
I get that oscillating between 1.2G and .6G [1200000000 and 600000000 ]
temp is 45.5'C approx
with nothing else running
the power light does not flash :-(

dang I'm going to have to fiddle more :(
It will clock down during idle time, is that what you are seeing?

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

Posted: Fri Mar 04, 2016 4:00 pm
by rurwin
That's the idle behaviour isn't it? When there is nothing to do the clock speed is dropped. Have you tried running at 100% load?

Edit: two minds with but a single thought, and both of them late to the party.

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

Posted: Fri Mar 04, 2016 4:04 pm
by RaTTuS
yes the previous post is when run
sysbench --num-threads=8 --test=cpu --cpu-max-prime=10000000000 run
on another terminal - I've updated that post

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

Posted: Sat Mar 05, 2016 2:19 pm
by DougieLawson
rurwin wrote:I thought it was defined by the skin you chose. The default skin (raspite) shows small pics, but the other (prosilver) shows them at full size.

In prosilver, if you right-click them and select "View Image", the browser will automatically scale it so it fills the window. It's still a pain to have to scroll past though.
Perhaps that will improve when raspite is moved to /dev/null and a new skin finally appears on the forum.

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

Posted: Sun Mar 06, 2016 11:14 am
by jeuf
Hi Guys,
They said we need a 2.5A power supply for the Pi3, but I guess it might be oversized, like it was on previous Pi
Pi2 draw only 400-500 mA with 4 core loaded for instance but it was advised to use a 1A power supply if I am not wrong.

Has anybody measured the current consumption of the Pi3, at idle and full load ? Thanks !

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

Posted: Sun Mar 06, 2016 11:23 am
by DougieLawson
Four cores @ 500mA plus four full power USB devices @ 500mA == one power supply @ 2500mA (2.5A)

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

Posted: Sun Mar 06, 2016 11:45 am
by jeuf
This is not enough accurate, sorry... I don't use USB devices, and especially I don't think the law is that simple. (500 mA x number of used core).
For instance on the Pi2 each additionnal core uses 50 mA

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

Posted: Sun Mar 06, 2016 11:53 am
by GTR2Fan
jeuf wrote:Pi2 draw only 400-500 mA with 4 core loaded for instance but it was advised to use a 1A power supply if I am not wrong.
If you had nothing pulling current from the USB ports and ignored the start-up surge current of the Pi2B, 1A may just about cut it. Allowing for a full 1.2A being drawn from the USB ports and start-up surge current, the recommendation for the Pi2B has been 2A for some time I believe. It's now 2.5A for the Pi3B for the same reasons.

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

Posted: Sun Mar 06, 2016 1:09 pm
by hippy
jeuf wrote:Has anybody measured the current consumption of the Pi3, at idle and full load ? Thanks !
https://www.raspberrypi.org/help/faqs - Pi 3B, Amps ...

Idle Avg 0.30
Stress Max 1.34

At 1.2A of USB draw and consumption would be 2.5A

If using a Pi Camera, Official LCD Display, it will be higher still.

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

Posted: Tue Mar 08, 2016 8:18 pm
by jeuf
Thanks a lot for your feedbacks !!

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

Posted: Tue Mar 08, 2016 9:17 pm
by 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.
img_thermal_1457442329126.jpg
without heatsink
img_thermal_1457442329126.jpg (44.98 KiB) Viewed 4093 times
After heatsink treatment: http://amzn.to/1Tqednr, temperature went down to a more manageable 50C.
img_thermal_1457442285477.jpg
with heatsink
img_thermal_1457442285477.jpg (50.56 KiB) Viewed 4093 times

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

Posted: Tue Mar 08, 2016 10:35 pm
by Heater
loverpi,

You know, looking at your thermal images what I see in both cases is a "white hot" area around the SoC.

Sure enough your "after" image shows a nice cool heat sink. But the area of PCB around the SoC is the same white hot.

The only other reference we have is at the bottom left of both images, 72F in one and 75F in the other. Basically the same.

I could conclude from this that the heat sink is not doing anything useful. The SoC and PCB around it are the same "white hot".

I do conclude that your methodology is flawed. A heat sink is not required. There is no "overheating issue" and your "internal engineering review" is fiction.

Basically SPAM.

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

Posted: Tue Mar 08, 2016 10:43 pm
by GTR2Fan
If you overlay the two images, the 'white hot' area on the one with the heatsink is actually larger than on the one with no heatsink. Doesn't this mean that the heatsink must be radiating the heat it absorbs back onto the board?

The product description describes dimensions of 0.8" x0.8" x0.8" (just over 20mm). If that's the case then the heatsink on the SOC has far too large a footprint so overspills the lid of the SOC. If that's the case then it's bound to radiate down onto the board as well as in other directions.

That's why the heatsink I chose is 14x14x10mm. It fits the SOC perfectly.

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

Posted: Tue Mar 08, 2016 10:50 pm
by mikronauts
The heatsink is shown to be cool, therefore it is not making good contact with the SOC.

If it were working properly, the white area around the CPU would be greatly reduced, and the heatsink would show up as being hotter than it currently shows.

(I used to do extreme overclocking on AMD/Intel processors...)

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

Posted: Tue Mar 08, 2016 10:58 pm
by loverpi
Heater wrote:loverpi,

You know, looking at your thermal images what I see in both cases is a "white hot" area around the SoC.

Sure enough your "after" image shows a nice cool heat sink. But the area of PCB around the SoC is the same white hot.

The only other reference we have is at the bottom left of both images, 72F in one and 75F in the other. Basically the same.

I could conclude from this that the heat sink is not doing anything useful. The SoC and PCB around it are the same "white hot".

I do conclude that your methodology is flawed. A heat sink is not required. There is no "overheating issue" and your "internal engineering review" is fiction.

Basically SPAM.
We were able to corrupt the HDMI steam by running sysbench with 4 threads in 22C ambient. SoC temperatures reached 100C via the thermal camera, infrared gun, and thermal probe. The second thermal image was taken without cooling down the board and 3 minutes of load with no active cooling. The contact probe reported 55C and the thermal image is inaccurate because of the different emissivity of aluminum.

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

Posted: Tue Mar 08, 2016 11:16 pm
by GTR2Fan
I can go along with the emissivity argument (even if you can't spell aluminium ;)), but how do you explain the 'white hot' area around the heatsink being larger than on the non-heatsinked board? Shouldn't the heatsink be radiating heat away from the board, not towards it?

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

Posted: Tue Mar 08, 2016 11:17 pm
by Heater
@loverpi,
We were able to corrupt the HDMI steam by running sysbench with 4 threads in 22C ambient.
I look forward to being able to repeat your experiment.
The second thermal image was taken without cooling down the board...
Did I say your methodology was flawed?

@ GTR2Fan
If you overlay the two images, the 'white hot' area on the one with the heatsink is actually larger than on the one with no heatsink...
That indicates to me that the heat sink is acting as more of a thermal insulator than a dissipater.

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

Posted: Tue Mar 08, 2016 11:37 pm
by Htaborda
I must say I'm quite disappointed with the thermal profile of the RPi3.

I recently got 2 RPi3, and I've replaced some previous instances where I was using the RPi2.
When I started checking the temperature, I noticed that my idle temperature is slightly above 50C. High for what I was used to, but not a problem in itself.

When put under load, the RPi3 shoots easily in the 70s C range... As an example, watching a 3gb HD movie meant cruising at 65C, and restoring a KODI backup made it reach 74C. Under a proper stress test, both units reach 85C, and start aggressively throttling down to 800Mhz. So essentially, I bought 2/3 of a RPi3.

I am well aware that it's not going to catch fire. The throttling is the expected behaviour, and 85C is well below the threshold that will damage any chip or component on the board.
However, this is not what I signed up for. As with the Rpi1 and RPi2, I was expecting to have a fully performing product without any gimmicks, throttling or exotic heatsink for a stock unit.

To compound the issue, I compared my results with 2 friends with very similar setups. They have not reached the 85C ceiling when running the same stress test. This means that the RPi3 is being produced with quite some variation between units. It feels that buying an RPi3 right now is more of a lottery. I feel like sending back my units.

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

Posted: Wed Mar 09, 2016 12:11 am
by loverpi
Heater wrote:@loverpi,
That indicates to me that the heat sink is acting as more of a thermal insulator than a dissipater.
The white area is only 50C due to the gradient issue I mentioned. There's thermal conduction from the bottom of the SoC to the board and the white should get progressively larger over time until it reaches the asymptote . There's no active cooling and the heatsinks reduce the SoC temperature by around 40C.

We use 3M thermal tape. You can get better cooling via thermal paste but that's precarious as there's no mounting hardware and your heatsink will fall off if it's not perfectly flat.

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

Posted: Wed Mar 09, 2016 7:06 am
by Heater
@Htaborda,

So you got WIFI and Bluetooth and multiple times the performance of previous Pi models, for the same price, and you are disappointed?

@loverpi
...the heatsinks reduce the SoC temperature by around 40C...
I would imagine the heat sink must do something.

To my simple mind if the source of heat, the SoC, were 40C cooler I'd expect the surrounding board area to be cooler as well. You thermal images do not show any such effect, The is the same "white hot" over the same area.

It does not add up. I might be convinced if you repeated the experiment, starting from a cold Pi in each case.

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

Posted: Wed Mar 09, 2016 7:10 am
by Pithagoros
Htaborda wrote:
To compound the issue, I compared my results with 2 friends with very similar setups. They have not reached the 85C ceiling when running the same stress test. This means that the RPi3 is being produced with quite some variation between units. It feels that buying an RPi3 right now is more of a lottery. I feel like sending back my units.
This is a useful paragraph. I was originally considering underclocking because of some of the reports on these forums, but in real use my Pi 3 hasn't actually got very warm at all. Monitoring temperature using the internal facility I am seeing 40-45C with no heatsink fitted, the Pi3 sitting on a desktop with no case doing normal coding work. I've been wondering why I've not seen any significant heating.

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

Posted: Wed Mar 09, 2016 7:28 am
by ssvb
Pithagoros wrote:This is a useful paragraph. I was originally considering underclocking because of some of the reports on these forums, but in real use my Pi 3 hasn't actually got very warm at all. Monitoring temperature using the internal facility I am seeing 40-45C with no heatsink fitted, the Pi3 sitting on a desktop with no case doing normal coding work. I've been wondering why I've not seen any significant heating.
Well, one of the concerns raised in this topic was about the internal sensor being not exactly accurate, as the topic name suggests. So the differences between individual units might be not only manifesting themselves as different levels of actual power consumption, but also as having a substantially different internal temperature sensor bias too.

In order to observe significant heating, you need to load the Raspberry Pi 3 board with some work stressing all CPU cores simultaneously. One of such examples is a parallel build of some C code using the GCC compiler. Please note that the GCC compiler is a relatively light workload though and much heavier workloads exist. For example, NEON optimized video transcoding from one format into another is rather power hungry. BTW, video playback is not so heavy because it does not need faster than realtime decoding and the CPU usually has a bit of time to rest between decoding individual frames.

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

Posted: Wed Mar 09, 2016 7:38 am
by ssvb
loverpi wrote: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.
This is a bit suspicious, because thermal throttling was expected to be kicking in and keeping the temperature reasonable. What was the internal sensor reporting at that time?

One more random guess is that GPU thermal throttling might be not implemented yet and CPU throttling alone might be insufficient. But if this is the case, then it should be fixable in software.