nooryani84
Posts: 34
Joined: Tue Aug 13, 2013 1:20 pm

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

Fri Mar 04, 2016 5:17 am

Wait.. so even with a heatsink it won't run at 1200 mhz under full load? That is disappointing indeed.

User avatar
Fidelius
Posts: 460
Joined: Wed Jan 01, 2014 8:40 pm
Location: Germany

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

Fri Mar 04, 2016 7:31 am

nooryani84 wrote:Wait.. so even with a heatsink it won't run at 1200 mhz under full load?
Yes, when the Pi is inside a case like my Pi2 Aukru case, which I linked to a few posts earlier.

My Pi2 case included a simple aluminum heatsink having the same size as the Broadcom ARM/GPU/RAM chip and isn't needed with a Pi2. The heatsink has on its bottom a black double-sided adhesive tape (i.e. no special heat paste). Still this simple heatsink takes away ~10° C from the Pi3 in "desktop idle mode".

However when I open the case, the Pi3's throttling which starts at around ~80°C, levels out at nearly 1200 MHz under full load, generated via the Debian command "stress" – i.e. for CPU only load. So I told my family to open the case when they want to include the Pi3 in their Blender rendering network. :-)

That is with the Videcore4 (VC4) being in idle mode, basically.
It would be interesting to see the temperature when the VC4 and the ARM have full load in their combined Broadcom chip.
Anybody? :-)

nooryani84
Posts: 34
Joined: Tue Aug 13, 2013 1:20 pm

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

Fri Mar 04, 2016 7:36 am

Ok, well I've ordered the official raspberry pi case. I hope that the inclusion of a heatsink will let me run it at 1200 mhz during full load.

ssvb
Posts: 112
Joined: Sat May 19, 2012 6:15 pm

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

Fri Mar 04, 2016 7:37 am

Fidelius wrote:
nooryani84 wrote:Whats the point with a 1200 Mhz clockspeed if it can only maintain it a few minutes?
Yes, this behavior is new for us Pi users. So far with our Pi1 and Pi2 we never needed to worry about any overheating.

So it's indeed a bit disappointing when you intend to use the Pi3 inside a case – you won't be able to have 1200 MHz under full-load then, even with passive cooling (heatsink). When I open the case however, the Pi3 levels out at a few decades below 1200 MHz.

Greetings to Norway!

P.S. Full load meaning something like:

Code: Select all

stress --cpu 4 --timout 600
Just some numbers for comparison. On a PINE64+ board (quad Cortex-A53 clocked at 1152MHz):

* running "sysbench --test=cpu --num-threads=4 run" results in a ~280mA current draw increase (compared to idle) from a 5V PSU.
* running "stress --cpu 4" results in a ~520mA current draw increase from a 5V PSU.
* running "minerd --benchmark" (CPU miner for Litecoin and Bitcoin) results in a ~980mA current draw increase from a 5V PSU.
* running "cpuburn-a53" results in a ~1650mA current draw increase from a 5V PSU.

As we can see, the "minerd" application is roughly twice more power hungry than your "stress" application. And this is a reasonably realistic workload (just one example of a decently NEON optimized code). The users may encounter such workloads in the wild.

The "cpuburn-a53" application is roughly three times more power hungry than the "stress" application and almost six (!) times more power hungry than "sysbench". But admittedly this is just a synthetic workload, which is unlikely to be encountered when running real applications :)

User avatar
Fidelius
Posts: 460
Joined: Wed Jan 01, 2014 8:40 pm
Location: Germany

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

Fri Mar 04, 2016 7:39 am

Pithagoros wrote: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?
You mean underclock constantly, I suppose. Then I don't know the answer.

When in idle mode, i.e. without CPU load, the Pi3's ARM8 is clocked with 600 MHz, and the Videocore with 250 MHz.
Under full CPU load this changes so:

Code: Select all

vcgencmd measure_clock arm
600,000,000 --> 1,200,000,000

Code: Select all

vcgencmd measure_clock core
250,000,000 --> 400,000,000

User avatar
Fidelius
Posts: 460
Joined: Wed Jan 01, 2014 8:40 pm
Location: Germany

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

Fri Mar 04, 2016 7:53 am

ssvb wrote:As we can see, the "minerd" application is roughly twice more power hungry than your "stress" application. And this is a reasonably realistic workload (just one example of a decently NEON optimized code). The users may encounter such workloads in the wild.

The "cpuburn-a53" application is roughly three times more power hungry than the "stress" application and almost six (!) times more power hungry than "sysbench". But admittedly this is just a synthetic workload, which is unlikely to be encountered when running real applications :)
Thanks for the numbers which are very interesting.
So a "minerd" or "cpuburn-a53" combined with some VC4 glbench2 would give interesting temperaturs on a Pi3, I suppose. :-)

ssvb
Posts: 112
Joined: Sat May 19, 2012 6:15 pm

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

Fri Mar 04, 2016 8:18 am

Fidelius wrote: Thanks for the numbers which are very interesting.
So a "minerd" or "cpuburn-a53" combined with some VC4 glbench2 would give interesting temperaturs on a Pi3, I suppose. :-)
Yes, Cortex-A53 is a true dual issue superscalar with a full width NEON unit. It has 2x faster peak performance per MHz compared to Cortex-A7 (on assembly optimized code) and can easily do heavy number crunching. But the peak power consumption is also higher. Doing something on the GPU at the same time is going to consume even more power.

The temperatures are not so interesting because there is throttling. It's all the same as the Intel turbo boost. What really matters is ensuring that the thermal throttling actually works reliable and the board is able to handle sudden spikes of power consumption gracefully.

nooryani84
Posts: 34
Joined: Tue Aug 13, 2013 1:20 pm

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

Fri Mar 04, 2016 8:34 am

So how aggressive is the throttling? How much does it scale down to when it reaches the temp limit? I'm wondering if I would have been better off with a Pine64 now

ssvb
Posts: 112
Joined: Sat May 19, 2012 6:15 pm

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

Fri Mar 04, 2016 9:00 am

nooryani84 wrote:So how aggressive is the throttling? How much does it scale down to when it reaches the temp limit? I'm wondering if I would have been better off with a Pine64 now
The A64 SoC in Pine64 is using the same 40nm manufacturing process as BCM2837 in Raspberry Pi 3, so the power consumption is likely comparable. The Pine64 board also does not have a heatsink out of the box just like Raspberry Pi 3 and throttles almost immediately to something around ~1GHz under heavy load. You would not be better off with a Pine64 board (assuming that RPi3 eventually gains the ability to run 64-bit kernels) because Pine64 is a cost optimized board and this clearly shows. In fact I would probably get a Raspberry Pi 3 board myself if it was available earlier :) You can find a comparison between the Raspberry Pi 3, ODROID-C2 and Pine A64+ boards here: http://www.cnx-software.com/2016/03/01/ ... comparison

Regarding the peak power consumption. The use of the Micro-USB connector for power is a bit questionable, because some people might be tempted to use a generic Micro-USB cable and connect it to their PC. And this is not a very good idea.

User avatar
Fidelius
Posts: 460
Joined: Wed Jan 01, 2014 8:40 pm
Location: Germany

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

Fri Mar 04, 2016 9:01 am

Ssvb, very interesting things you tell! The Pi3's ARM8 Cortex-A53 really starts to fascinate me. I wouldn't have imagined 20 years ago when we hand-coded the ARM2,3,6 and StrongARM in our Acorn Archimedes'es, that one day we would have such a "beast" under our fingers – let alone taking the size of a 3½'' disc...
nooryani84 wrote:So how aggressive is the throttling? How much does it scale down to when it reaches the temp limit?
Soft but reliable throttling-back according to my simple tests. I.e. no saw tooth. Looks to be efficient.

User avatar
RaTTuS
Posts: 10548
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

Fri Mar 04, 2016 9:05 am

I think there are a few issues here -
one if you run the GPU / CPU cores flat out 24x7 in a case then you probably do
if oyu overclock you probably do
there may be an issue with the on demand cpu scaler not switching correctly
there can and probably will be fixed
http://uk.farnell.com/amec-thermasol/mp ... dp/1892471 < may well be the best bet for heatsinks

but the best advice currently is

DON'T PANIC
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

nooryani84
Posts: 34
Joined: Tue Aug 13, 2013 1:20 pm

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

Fri Mar 04, 2016 9:10 am

RaTTuS wrote:I think there are a few issues here -
one if you run the GPU / CPU cores flat out 24x7 in a case then you probably do
if oyu overclock you probably do
there may be an issue with the on demand cpu scaler not switching correctly
there can and probably will be fixed
http://uk.farnell.com/amec-thermasol/mp ... dp/1892471 < may well be the best bet for heatsinks

but the best advice currently is

DON'T PANIC
Thanks. You're probably right on the last part :) I'm sure that it will perform well, I was just worrying about sudden/dramatic decrease in performance when playing a game or watching a movie. I guess I'm making a mountain out of a molehill

I'm picking it up today and very excited to see how it performs. Haven't touched anything but the first Pi :)

User avatar
RaTTuS
Posts: 10548
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

Fri Mar 04, 2016 9:21 am

nooryani84 wrote:...or watching a movie. I guess I'm making a mountain out of a molehill

I'm picking it up today and very excited to see how it performs. Haven't touched anything but the first Pi :)
if your movies are h264 then without any codecs you can watch with no issues - 256Mb Pi 1 will do that
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

nooryani84
Posts: 34
Joined: Tue Aug 13, 2013 1:20 pm

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

Fri Mar 04, 2016 9:24 am

What if I'm streaming H.265? I would hope that this will work fine as well if I have a heatsink.

User avatar
RaTTuS
Posts: 10548
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

Fri Mar 04, 2016 9:30 am

I don't have much experience with h265 though reports here say most work on PI2's
higher bitrate ones may have issues - but those work fine on PI3's [or will]
if they have issues then it will be software and fixable in the short / medium term
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

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 25914
Joined: Sat Jul 30, 2011 7:41 pm

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

Fri Mar 04, 2016 9:57 am

nooryani84 wrote:What if I'm streaming H.265? I would hope that this will work fine as well if I have a heatsink.
Should be OK.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed. Here's an example...
“My wife said to me `...you’re not even listening`.
I thought, that’s an odd way to start a conversation.."

User avatar
Trixster
Posts: 261
Joined: Sat Jul 07, 2012 3:53 pm

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

Fri Mar 04, 2016 10:10 am

These are my findings so far.

My rpi2 was running at arm 1050mhz, core 500mhz, sdram 600mhz, schmoo 0x2000020

On medium bitrate 1080p x265 it would begin to stutter during action scenes. Slow scenes would be OK.

On the first run of a video with the rpi3 running at arm 1200mhz, core 500mhz, sdram 600, schmoo 0x2000020, medium bitrate 1080p x265 recordings do not stutter at all, even in high action scenes. And this is even the case when the temperature increases above 80degs and the core starts to throttle back down to below 950mhz. Force_turbo is 1.

However, during repeated runs of complex scenes (iron man vs hulk in the 2nd Avengers film for instance) with a pi that's been running videos for hours, I've seen the pi throttle back the arm right down to 600mhz and even the core being throttled back to 250mhz. In these instances then the 1080p x265 decode will occasionally slow and stutter. But as soon as a quieter scene comes along then the temps drop a little and the arm will go back up to the 1000-1200 range and playback becomes faultless again.

That's a pi3 in a pibow case with the lid off but no heatsink at the moment, i have some on the way.

User avatar
GTR2Fan
Posts: 1601
Joined: Sun Feb 23, 2014 9:20 pm
Location: South East UK

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

Fri Mar 04, 2016 10:15 am

nooryani84 wrote:I was just worrying about sudden/dramatic decrease in performance when playing a game...
I suggest that somebody installs Quake 3 Arena on a Pi3B to check this out...

viewtopic.php?f=78&t=18853

...as this has my enclosed and overclocked Pi2B (see sig) extremely close to thermal-throttling with an ambient temperature of just 17°C.
Pi2B Mini-PC/Media Centre: ARM=1GHz (+3), Core=500MHz, v3d=500MHz, h264=333MHz, RAM=DDR2-1200 (+6/+4/+4+schmoo). Sandisk Ultra HC-I 32GB microSD card on '50=100' OCed slot (42MB/s read) running Raspbian/KODI16, Seagate 3.5" 1.5TB HDD mass storage.

hippy
Posts: 7103
Joined: Fri Sep 09, 2011 10:34 pm
Location: UK

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

Fri Mar 04, 2016 10:45 am

RaTTuS wrote:DON'T PANIC
That would seem to be the best advice.

I would guess the Pi 3 development team never expected or predicted the few hiccups we are seeing with heating and other changes and I think we can be confident they will be investigating those, gaining a better understanding of any issues and their scope, looking for solutions, mitigations and workarounds.

It is interesting that boards seem to behave differently under loading, some not heating as highly as others. I am sure they will be intrigued by that as much as we all are. I have my own theory on that but it would only amount to concern trolling so I will keep that to myself and let those who designed the boards get on with investigating what issues there may be.

JimmyN
Posts: 1109
Joined: Wed Mar 18, 2015 7:05 pm
Location: Virginia, USA

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

Fri Mar 04, 2016 12:08 pm

If you want to compare different boards to look for variance you need to use exactly the same stress test on each, or it's not a true comparison. Just because the CPU is at 100% doesn't necessarily mean it's at peak power, 100% clock doesn't necessarily mean 100% power.

As an example the PC I'm typing this on is running an i7 2700K+ CPU overclocked to 5.1gHz. It's a 95W TDP processor. If it's doing ray tracing, or running filters in Photoshop the CPU will be at 100% and TDP will be around 95W, so you could say it's drawing max power and dissipating max heat, but it's not.

Run Prime95 and the CPU is at 100% but the TDP jumps to 125W, up 30W from it's rating so it must be at max power now, but it's not. Run Linpack, Intel's burn test and the TDP goes all the way up to 185W, almost double it's power rating and liquid cooling is needed at that point.

So a "100%" CPU load on this system can be 95W, or all the way up to 185W, depending on what is being used to test it. If two people are running stress tests on the RPi and one sees 60C and another sees 85C it doesn't mean anything unless they are both using the same stress test.

User avatar
GTR2Fan
Posts: 1601
Joined: Sun Feb 23, 2014 9:20 pm
Location: South East UK

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

Fri Mar 04, 2016 12:11 pm

JimmyN wrote:If two people are running stress tests on the RPi and one sees 60C and another sees 85C it doesn't mean anything unless they are both using the same stress test.
+1
Pi2B Mini-PC/Media Centre: ARM=1GHz (+3), Core=500MHz, v3d=500MHz, h264=333MHz, RAM=DDR2-1200 (+6/+4/+4+schmoo). Sandisk Ultra HC-I 32GB microSD card on '50=100' OCed slot (42MB/s read) running Raspbian/KODI16, Seagate 3.5" 1.5TB HDD mass storage.

User avatar
chrisoh
Posts: 217
Joined: Sun Dec 06, 2015 8:50 pm
Location: Essex, UK

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

Fri Mar 04, 2016 12:19 pm

GTR2Fan wrote:
JimmyN wrote:If two people are running stress tests on the RPi and one sees 60C and another sees 85C it doesn't mean anything unless they are both using the same stress test.
+1
This is what Gareth is undertaking here: https://www.reddit.com/r/raspberry_pi/c ... _heat_and/
Raspberry Pi 3 model B Idle @ 250mA
  • HDMI +200mA
  • Minecraft 1.9 Server +100mA
  • 2.4Ghz dongle +20mA

User avatar
RaTTuS
Posts: 10548
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

Fri Mar 04, 2016 12:43 pm

my RPI issue
viewtopic.php?f=63&t=138193&start=25#p918968
seems to be a bad PSU ...
which is weird as that was a previously good PSU / Micro USB lead, however I do get the power light flashing
more tests
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
rurwin
Forum Moderator
Forum Moderator
Posts: 4258
Joined: Mon Jan 09, 2012 3:16 pm
Contact: Website

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

Fri Mar 04, 2016 1:48 pm

My works-with-other-Pi's power supply was horrible with the Pi3. It didn't even get through a boot. The old RS supplied one seemed to work OK but I may have had the odd flash on the power LED. I've upgraded now to the newer 5.1V, 2.5A version.

User avatar
RaTTuS
Posts: 10548
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

Fri Mar 04, 2016 2:07 pm

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 :(
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

Return to “General discussion”