ejolson
Posts: 5800
Joined: Tue Mar 18, 2014 11:47 am

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

Sat Sep 10, 2016 8:57 pm

DavidS wrote:running an RPi 2B at 1100MHz (1.1GHz) for a long time, and often doing things that push all of the ARM cores at 100% for hours at a time, I have never had the tempurature exceed 49 degrees C.
It is curious and possibly the point of this thread that the Pi 3 generates significantly more heat than the 2B. At least the Pi 3 CPU isn't right next to a lithium-ion battery that might explode when things overheat, such as happens with unlucky tablets and smart phones.

dpotop
Posts: 86
Joined: Mon Nov 24, 2014 2:14 pm

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

Sat Sep 10, 2016 9:31 pm

ejolson wrote:
dpotop wrote:Small contribution to this thread: I also noticed a discrepancy between actual and measured temperature, but only after installing a heat sink on the SoC. Here's how it worked:
- At first, I used my RPi3 without a heat sink. During heavy loads (e.g. compilation) temperature went to 80C and then stabilized with +/-1 degree variations (the expected behavior). When temperature increased, frequency would go down, and the reverse.
- Once I installed the heat sink (aluminium, taken from the chipset of an old PC motherboard) the temperature stabilized around 70 degrees, and there was some throttling.

So I'm guessing that there are 2 temperature sensors:
- One is used by the CPU to detect overheating and to throttle. I would be curious to know the cut-off temperature.
- The second is somewhere else in the SoC casing, and it can be read in software.
When adding a heat sink, the temperature gradient changes.

Let me know if you have more info on this.

Best,
Dumitru
Are you measuring temperature with an external sensor or probe of some sort?
No. I'm just using the internal sensor.
dpotop

Umino
Posts: 5
Joined: Thu Aug 25, 2016 7:02 pm

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

Sat Sep 10, 2016 9:37 pm

I would even go as far as saying that an all-passive cooling does not cut it, not even with a heatsink on.

when I ordered my pi 3, I bought a heatsink kit with it and installed it right after unboxing.

first OS I tested on the pi was ubuntu mate and there it happened - while doing apt upgrade I noticed a little square showing in the upper right corner of the screen after a while - I googled that up and found out it is either indicating a possible problem with the power if the square is rainbow colored or an overheat indicator if it is all red - it wasn't rainbow colored, but changed color from orange to red - shortly after hitting red color the pi crashed.

I still had the case open by then so a lack of ventilation in the case could not have been the reason for this.

Then I took a 40mm fan wich I had around, laid it on the pi board and connected it to 5V - apt upgrade did crash again but the square didn't show up this time.

then I opened youtube and tried some 720p videos because that produces a lot of CPU load - I did this several times with and without the fan running and the orange/red square always started to show after a while when the fan was off ... it never showed when it was on.

so my recommendation is: when you put heavy load to your pi 3 then a simple heatsink and good ventilated case alone is NOT sufficient.

now I have a 30mm fan installed in my case and wired to 3.3v to keep it silent - that seems to be enough

dom
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 5592
Joined: Wed Aug 17, 2011 7:41 pm
Location: Cambridge

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

Mon Sep 12, 2016 12:47 pm

Umino wrote:I would even go as far as saying that an all-passive cooling does not cut it, not even with a heatsink on.
Depends how good the passive cooling is. A flirc case doesn't require a fan and it won't throttle even when stressed.
Effectively the whole case is the heat sink and it only gets moderately warm.

2badmice
Posts: 56
Joined: Thu Nov 03, 2016 9:21 pm
Location: North West England

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

Mon Dec 05, 2016 2:11 pm

rurwin wrote: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.

Thanks for this reply :)
Pi Noob, please be gentle

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 27061
Joined: Sat Jul 30, 2011 7:41 pm

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

Mon Dec 05, 2016 3:52 pm

DavidS wrote:This seems to negate one of the major advantages of using an ARM. That is the fact that the ARM always runs cool.

If it is really getting over 50 degrees C, then it should be rated to run at what ever clock it keeps down under 50 degrees C with a 100% load for an extended time on all 4 ARM cores.

running an RPi 2B at 1100MHz (1.1GHz) for a long time, and often doing things that push all of the ARM cores at 100% for hours at a time, I have never had the tempurature exceed 49 degrees C.
ARM have an advantage because they are not as powerful as something like x86. If you had an ARM with the same performance as a x86 on the same node, then the ARM would also run hot, and require similar levels of power. Theoretically, at least. That doesn't take in to account how well the architecture is implemented but since both ARM and Intel know how to make chips, then I would presume that both are well implemented.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed.
I've been saying "Mucho" to my Spanish friend a lot more lately. It means a lot to him.

2badmice
Posts: 56
Joined: Thu Nov 03, 2016 9:21 pm
Location: North West England

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

Mon Dec 05, 2016 4:49 pm

dom wrote:
Umino wrote:I would even go as far as saying that an all-passive cooling does not cut it, not even with a heatsink on.
Depends how good the passive cooling is. A flirc case doesn't require a fan and it won't throttle even when stressed.
Effectively the whole case is the heat sink and it only gets moderately warm.


I'm running a Plex Media Server on my Pi3, not sure if heat is an issue for me, but that case looks good. Shame it's not in the UK
Pi Noob, please be gentle

Heater
Posts: 16545
Joined: Tue Jul 17, 2012 3:02 pm

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

Mon Dec 05, 2016 5:50 pm

jamesh,
ARM have an advantage because they are not as powerful as something like x86.
Back in the day of the Acorn RISC machine the ARM not only consumed far less power than x86 of the time but was much faster.

Things moved on of course. Intel went for maximum performance in your PC. With long pipelines, bigger caches all kind of speed ups that require ever more transistors and more power. ARM landed in mobile devices where saving battery power was the name of the game and raw performance was secondary.

So what is the situation today? Is there an Intel x86 based system, like the Edisson or whatever, that has performance and power consumption that matches the Pi 3 or other ARM systems?

Is there an Intel system that matches the likes of the STM32 F4 in performance and power consumption?
Memory in C++ is a leaky abstraction .

MarkTF
Posts: 318
Joined: Tue Mar 03, 2015 4:59 pm

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

Mon Dec 05, 2016 6:29 pm

Heater wrote:jamesh,
So what is the situation today? Is there an Intel x86 based system, like the Edison or whatever, that has performance and power consumption that matches the Pi 3 or other ARM systems?

Is there an Intel system that matches the likes of the STM32 F4 in performance and power consumption?
The Intel Atom line of Intel x86 processors is nominally the competitor to Pi 3 class ARM processors. I have, for instance, a cheap 7" tablet with an Atom processor (Z3735G - 1.8 GHz quad core). I don't think Intel has been particularly successful in the tablet/phone market, but there are a lot of laptops using Atoms.

Quark is Intel's x86 microcontroller line spanning a similar range of performance as the STM32 line. Intel is pretty new to this market space. This is the processor that is used in the Intel Edison module and Galileo board (400 MHz single core).

Heater
Posts: 16545
Joined: Tue Jul 17, 2012 3:02 pm

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

Mon Dec 05, 2016 6:40 pm

Intel has been in this market space before. Remember the 80186? Remember the i960? They were even building ARM chips at one point.

Obviously things are bit different today. Hence my question, is there a system from Intel that matches P3 performance and power consumption?
Memory in C++ is a leaky abstraction .

MarkTF
Posts: 318
Joined: Tue Mar 03, 2015 4:59 pm

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

Mon Dec 05, 2016 8:41 pm

Heater wrote:Obviously things are bit different today. Hence my question, is there a system from Intel that matches P3 performance and power consumption?
The aforementioned Atom Z3735G SOC is nominally 2.2 W in a tablet application where one more commonly finds a quad core ARM of the Pi3 class, probably a bit more power consumption and computing power than the Pi3, but in the same ballpark.

There exist a few "hacker boards" using these processors, but I don't have hands on experience. I think they all come with passive heat sinks (consistent with higher typical power) and they're more expensive than an Pi3.

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 27061
Joined: Sat Jul 30, 2011 7:41 pm

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

Mon Dec 05, 2016 8:45 pm

Heater wrote:jamesh,
ARM have an advantage because they are not as powerful as something like x86.
Back in the day of the Acorn RISC machine the ARM not only consumed far less power than x86 of the time but was much faster.

Things moved on of course. Intel went for maximum performance in your PC. With long pipelines, bigger caches all kind of speed ups that require ever more transistors and more power. ARM landed in mobile devices where saving battery power was the name of the game and raw performance was secondary.

So what is the situation today? Is there an Intel x86 based system, like the Edisson or whatever, that has performance and power consumption that matches the Pi 3 or other ARM systems?

Is there an Intel system that matches the likes of the STM32 F4 in performance and power consumption?
There is a fundamental rule here - at the same process node, for the same power consumption, it doesn't matter what the architecture is, the processor power with be roughly similar for similar die areas IF the implementation quality is similar (e.g. caching pipelining, prediction etc). Nowadays, the underlying micro-architecture implements the instruction set, for example X86 is not implemented directly, but over a micro-architecture, which makes it pretty much the same as ARM.

Of course, Intel use much smaller process nodes, and many more transistors, so they can perform much better, but also use more power. Remember, when it comes down to it, processors are mainly transistors, and transistors take similar levels of power whatever the architecture. So, a combination of die area and process node gives the total number of transistors. And that is what determines overall power and power consumption.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed.
I've been saying "Mucho" to my Spanish friend a lot more lately. It means a lot to him.

ag123
Posts: 111
Joined: Sun Dec 18, 2016 7:54 am

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

Thu Dec 22, 2016 8:45 am

i'm trying to cool my hot pi 3 too :lol:

i think much of the higher power consumption is that we transits to superscalar processor could be a large factor. it possibly means *a large number more* transistors and that those transistors are all running to make super-scalar possible. hence, it *draws more power* and run *hotter* just as with the transition of the 'low end' intel atoms to those highly superscalar, out-of-order execution, parallel hyper threaded cores on the 'high end' intel i7 processors.

A53 is superscalar but in-order execution not sure when an A57 superscalar, out-of-order execution multi core would make reality on the R-pi, that'd be fun to push the boundaries of *gflops* and so is the fun of transiting to and joining the league of using monster (large) elaborate, over designed *cpu coolers* and overclocking r-pi to try to compete with supercomputers :lol:

has any 1 bothered to cool their pi 3 with liquid nitrogen yet ? not sure how many ghz can 1 attempt if that's done :lol:

RinksCustoms
Posts: 5
Joined: Sun Mar 26, 2017 9:08 pm

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

Sun Mar 26, 2017 9:41 pm

The discrepency error in temperature measurement is from multiple sources of error:
  • Diodes as temperature measuring devices while linear usually exhibit offsets and require precision calibration to be useful
  • internal Vref's that are likely used also exhibit offsets and would likely need calibration for precision measurements
  • internal temps will ALWAYS be higher due to substrate to case thermal resistance and case to ambient resistance
My Pi 3 has a Broadcom CPU. Sticky backed Al sink was warped and made lousy contact with the SOC (only the edges). Stress testing resulted in thermal throttling at 80C reported down to 900MHz @ 82C

I cut a copper slug about the size of the SOC and used Ghelid GC Extreeme in between my sandwich of the SOC/5mm thick copper bar/old finned Al sink i modded to fit over the SOC & copper piece. The Al sink is ~3.5x2.5x1cm.

Image
Image

Stress tested with --cpu 4 again and temp never hit 66C

Return to “General discussion”