User avatar
rpdom
Posts: 14722
Joined: Sun May 06, 2012 5:17 am
Location: Chelmsford, Essex, UK

Re: Heatsinks official for rpi 4 ?

Sat Jul 06, 2019 7:36 am

Ok, just for fun I decided to run a test on my Pi 4B.

I got an old PC CPU heatsink. Removed the fan and cut off a 10mmx10mm piece from it. I cleaned that up, put some cheap heatsink compound on it, and stuck it on top of the SoC (small blob of superglue on each corner to hold it in place).

I then ran "stress" for 15 minutes on all four CPU cores.

At 23°C ambient the Pi was idling at 51°C.
Under stress test it peaked at 76°C.

Not bad for a bit of a bodge. I might cut down the fins on the heatsink to remove some height. It's currently about 28mm high.

User avatar
HawaiianPi
Posts: 4453
Joined: Mon Apr 08, 2013 4:53 am
Location: Aloha, Oregon USA

Re: Heatsinks official for rpi 4 ?

Sat Jul 06, 2019 7:56 am

Gavinmc42 wrote:
Thu Jul 04, 2019 4:39 am
What is the smallest fan/heatsink combo available?
I have a 25mm fan around here somewhere, and a 25mm heatsink I cut down to go with it.

I think I posted a pic of it before, lemme look... Here ya go.
RPi_small_HSF.jpg
RPi_small_HSF.jpg (61.57 KiB) Viewed 1235 times

I should find it and try it on my Pi4 (when it arrives). Looks like it's on GF's Pi2, so I think I know where it is...

Oh, I have some old Tennmax Lasagna CPU Coolers as well. Bet that would work great (but I think they're 12V). Added to find list.
My mind is like a browser. 27 tabs are open, 9 aren't responding,
lots of pop-ups...and where is that annoying music coming from?

jahboater
Posts: 4595
Joined: Wed Feb 04, 2015 6:38 pm

Re: Heatsinks official for rpi 4 ?

Sat Jul 06, 2019 8:21 am

gennargiu wrote:
Sat Jul 06, 2019 6:33 am

[email protected]:~ $ while true; do vcgencmd measure_clock arm; vcgencmd measure_temp; sleep 10; done& stress -c 4 -t 100s
[1] 1507
stress: info: [1508] dispatching hogs: 4 cpu, 0 io, 0 vm, 0 hdd
Try cpuburn instead of stress. It will raise the temperature much higher (and quicker). It uses NEON on all four cores.

Code: Select all

wget https://raw.githubusercontent.com/ssvb/cpuburn-arm/master/cpuburn-a53.S
gcc -o cpuburn-a53 cpuburn-a53.S
./cpuburn-a53
Leave it running for an hour to stress test a CPU/heat sink. This little script is handy for monitoring the temps and any throttling (non zeros in health).

Code: Select all

#!/bin/bash
vcgencmd get_config int
for src in h264 isp uart pwm emmc pixel vec hdmi dpi ; do \
  echo -e "$src:\t$(vcgencmd measure_clock $src)" ; \
done
Counter=14
DisplayHeader="Time       Temp     CPU     Core     GPU          Health          Vcore"
while true ; do
  let ++Counter
  if [ ${Counter} -eq 15 ]; then
    echo -e "${DisplayHeader}"
    Counter=0
  fi
  Health=$(perl -e "printf \"%19b\n\", $(vcgencmd get_throttled | cut -f2 -d=)")
  Temp=$(vcgencmd measure_temp | cut -f2 -d=)
  Clockspeed=$(vcgencmd measure_clock arm | awk -F"=" '{printf ("%0.0f",$2/1000000); }' )
  Corespeed=$(vcgencmd measure_clock core | awk -F"=" '{printf ("%0.0f",$2/1000000); }' )
  GPUspeed=$(vcgencmd measure_clock v3d | awk -F"=" '{printf ("%0.0f",$2/1000000); }' )
  CoreVolt=$(vcgencmd measure_volts | cut -f2 -d= | sed 's/000//')
  echo -e "$(date '+%H:%M:%S')  ${Temp}  $(printf '%4s' ${Clockspeed})MHz $(printf '%4s' ${Corespeed
  sleep 5
done

chipace
Posts: 56
Joined: Sat Jun 29, 2019 2:56 am

Re: Heatsinks official for rpi 4 ?

Sat Jul 06, 2019 4:01 pm

Thanks jahboater. I ran cpuburn-a53 and the temps came up much faster. I didn't hit 75C after 5min (just showing the first 2min).

[email protected]:~ $ while true; do vcgencmd measure_clock arm; vcgencmd measure_temp; sleep 10; done& ./cpuburn-a53
[1] 1300
  • frequency(48)=600169920
    temp=43.0'C
    frequency(48)=1500345728
    temp=52.0'C
    frequency(48)=1500345728
    temp=54.0'C
    frequency(48)=1500345728
    temp=55.0'C
    frequency(48)=1500345728
    temp=58.0'C
    frequency(48)=1500345728
    temp=60.0'C
    frequency(48)=1500398464
    temp=61.0'C
    frequency(48)=1500398464
    temp=63.0'C
    frequency(48)=1500398464
    temp=64.0'C
    frequency(48)=1500345728
    temp=66.0'C
    frequency(48)=1500345728
    temp=67.0'C
    frequency(48)=1500398464
    temp=68.0'C
    frequency(48)=1500398464
    temp=68.0'C
    frequency(48)=1500345728
    temp=70.0'C
    frequency(48)=1500345728
    temp=71.0'C
    frequency(48)=1500345728
    temp=73.0'C

User avatar
5t4n5
Posts: 13
Joined: Mon Jul 01, 2019 5:22 pm
Location: Near a beach in Devon
Contact: Website Twitter

Re: Heatsinks official for rpi 4 ?

Mon Jul 08, 2019 7:15 am

I'm currently using one of these little 13mm things in black... https://thepihut.com/collections/raspbe ... i-heatsink

In this case.... https://www.ebay.co.uk/itm/Raspberry-Pi ... 2701060051

With the little tiny 30mm fan blowing down onto the heatsink.

Blasting the cpu with sysbench for 10 minutes and it doesn't go over 57, which i don't think is too bad for such a little heatsink and fan as a temporary solution, if not a permanent one.

[email protected]:~/test $ sudo ./test2.sh
temp=44.0'C
temp=55.0'C
temp=55.0'C
temp=56.0'C
temp=56.0'C
temp=57.0'C
temp=56.0'C
temp=56.0'C
temp=57.0'C
temp=57.0'C
temp=57.0'C
temp=56.0'C
temp=57.0'C
temp=57.0'C
temp=56.0'C
temp=57.0'C
temp=56.0'C
temp=56.0'C
temp=55.0'C
temp=56.0'C
temp=57.0'C
[email protected]:~/test $


I've got a bigger solution coming from China, will report back when it arrives and i get some results.
Honestly, i've no idea what i'm doing. :oops:

jahboater
Posts: 4595
Joined: Wed Feb 04, 2015 6:38 pm

Re: Heatsinks official for rpi 4 ?

Mon Jul 08, 2019 7:57 am

I use this heat sink from the PiHut which is bigger, heavier, and has decent 3M thermal tape pre-applied.
https://thepihut.com/products/al-heat-s ... 30-30-10mm
There is no fan, they are simply not needed for the Pi4 (unless you stick it in a small plastic case).
My Pi4 is overclocked and it still remains reasonably cool. Sysbench is a light load for Pi4.
Here are the temps and frequencies for sysbench run:

Code: Select all

Time       Temp     CPU     Core     GPU          Health          Vcore
08:46:46  39.0'C   400MHz  250MHz  250MHz  00000000000000000000  0.8455V
08:46:56  40.0'C   400MHz  250MHz  250MHz  00000000000000000000  0.8507V
08:47:07  38.0'C   400MHz  250MHz  250MHz  00000000000000000000  0.8507V
08:47:17  41.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:47:27  43.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:47:37  43.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:47:47  42.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:47:57  42.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:48:07  43.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:48:17  44.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:48:27  45.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:48:37  45.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:48:47  45.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:48:57  45.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:49:07  45.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
Time       Temp     CPU     Core     GPU          Health          Vcore
08:49:17  46.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:49:27  47.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:49:37  46.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:49:48  47.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:49:58  46.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:50:08  48.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:50:18  48.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:50:28  47.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:50:38  47.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:50:48  48.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:50:58  48.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:51:08  48.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:51:18  46.0'C   400MHz  250MHz  250MHz  00000000000000000000  0.8507V
08:51:28  46.0'C   400MHz  250MHz  250MHz  00000000000000000000  0.8507V
08:51:39  46.0'C   400MHz  250MHz  250MHz  00000000000000000000  0.8507V

RGJ-PI
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 20
Joined: Fri May 24, 2019 2:10 pm
Location: Cambridge UK

Re: Heatsinks official for rpi 4 ?

Mon Jul 08, 2019 11:13 am

If size isn't an issue but absolute performance is then this active solution would take some beating..
https://www.seeedstudio.com/ICE-Tower-C ... -4097.html
I think it's safe to say this isn't compatible with the official Pi 4 case!

cruster
Posts: 98
Joined: Mon Sep 01, 2014 7:56 pm

Re: Heatsinks official for rpi 4 ?

Mon Jul 08, 2019 11:38 am

I've been using this -

Code: Select all

https://uk.farnell.com/amec-thermasol/mpc222225t/heat-sink-ceramic-22-22-2-5-std/dp/1892474

ejolson
Posts: 3231
Joined: Tue Mar 18, 2014 11:47 am

Re: Heatsinks official for rpi 4 ?

Mon Jul 08, 2019 1:52 pm

jahboater wrote:
Mon Jul 08, 2019 7:57 am
I use this heat sink from the PiHut which is bigger, heavier, and has decent 3M thermal tape pre-applied.
https://thepihut.com/products/al-heat-s ... 30-30-10mm
There is no fan, they are simply not needed for the Pi4 (unless you stick it in a small plastic case).
My Pi4 is overclocked and it still remains reasonably cool. Sysbench is a light load for Pi4.
Here are the temps and frequencies for sysbench run:

Code: Select all

Time       Temp     CPU     Core     GPU          Health          Vcore
08:46:46  39.0'C   400MHz  250MHz  250MHz  00000000000000000000  0.8455V
08:46:56  40.0'C   400MHz  250MHz  250MHz  00000000000000000000  0.8507V
08:47:07  38.0'C   400MHz  250MHz  250MHz  00000000000000000000  0.8507V
08:47:17  41.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:47:27  43.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:47:37  43.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:47:47  42.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:47:57  42.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:48:07  43.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:48:17  44.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:48:27  45.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:48:37  45.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:48:47  45.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:48:57  45.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:49:07  45.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
Time       Temp     CPU     Core     GPU          Health          Vcore
08:49:17  46.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:49:27  47.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:49:37  46.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:49:48  47.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:49:58  46.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:50:08  48.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:50:18  48.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:50:28  47.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:50:38  47.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:50:48  48.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:50:58  48.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:51:08  48.0'C  1600MHz  600MHz  250MHz  00000000000000000000  0.8507V
08:51:18  46.0'C   400MHz  250MHz  250MHz  00000000000000000000  0.8507V
08:51:28  46.0'C   400MHz  250MHz  250MHz  00000000000000000000  0.8507V
08:51:39  46.0'C   400MHz  250MHz  250MHz  00000000000000000000  0.8507V
While sysbench sure seems useless as a stress test, it at least checks its computation of primes for correctness. While cpuburn-a53 raises the temperatures nicely, it contains no code to check the consistency of the computation that guards against processor malfunctions. What reasonable stress test is there?

jahboater
Posts: 4595
Joined: Wed Feb 04, 2015 6:38 pm

Re: Heatsinks official for rpi 4 ?

Mon Jul 08, 2019 2:13 pm

ejolson wrote:
Mon Jul 08, 2019 1:52 pm
What reasonable stress test is there?
In the past I have tried a combination of memtester and cpuburn in multiple terminal windows. Cpuburn gets the temperature up but does nothing with memory and as you say, doesn't check anything. A slightly harder test was multiple instances of memtester (say testing 25% of the free memory each simultaneously) and then combining that with cpuburn.

The trouble is with memtester on the Pi4 4GB is that it takes ages to check that larger amount of memory.

In the early days of the Pi3 it didn't matter that cpuburn didn't do any checks, simply because the Pi3 would crash after a few seconds anyway!

Now with the Pi4 I can run cpuburn indefinitely and it doesn't even throttle (gets to around 76-77C) even for my overclocked Pi4. 28nm is wonderful!

I suspect if you modified cpuburn to check things it might not run so hot.

User avatar
piglet
Posts: 900
Joined: Sat Aug 27, 2011 1:16 pm

Re: Heatsinks official for rpi 4 ?

Mon Jul 08, 2019 3:01 pm

RichardJones wrote:
Mon Jul 08, 2019 11:13 am
If size isn't an issue but absolute performance is then this active solution would take some beating..
https://www.seeedstudio.com/ICE-Tower-C ... -4097.html
I think it's safe to say this isn't compatible with the official Pi 4 case!
What's holding that on!?

ejolson
Posts: 3231
Joined: Tue Mar 18, 2014 11:47 am

Re: Heatsinks official for rpi 4 ?

Mon Jul 08, 2019 3:16 pm

jahboater wrote:
Mon Jul 08, 2019 2:13 pm
I suspect if you modified cpuburn to check things it might not run so hot.
Strangely enough I'm trying to sell licenses for a patent that uses the theory of deterministic chaos in nonlinear dynamical systems to get around the difficulty that checking the answer ruins the stress test. I think a nonexclusive license would be cheap, but one would need to work the details out through the technology transfer office. On the other hand, sometimes it's better to spend the afternoon in the garden pulling weeds.

The calculation which illustrated the difficulties with 3V power on the Pi 3B most dramatically, in my opinion, was the high-performance Linpack available here. The reported speed of the 4B is about 10.9 Gflops using the power-over-Ethernet hat as the cooling solution. While that number seems reasonable, an independent confirmation would be nice.

jahboater
Posts: 4595
Joined: Wed Feb 04, 2015 6:38 pm

Re: Heatsinks official for rpi 4 ?

Mon Jul 08, 2019 4:29 pm

ejolson wrote:
Mon Jul 08, 2019 3:16 pm
The calculation which illustrated the difficulties with 3V power on the Pi 3B most dramatically, in my opinion, was the high-performance Linpack available here. The reported speed of the 4B is about 10.9 Gflops using the power-over-Ethernet hat as the cooling solution. While that number seems reasonable, an independent confirmation would be nice.
I seem to remember in the past, there was a link to a pre-built executable?

The build fails and I would rather be watching Wimbledon that trying to correct the header file errors!

ejolson
Posts: 3231
Joined: Tue Mar 18, 2014 11:47 am

Re: Heatsinks official for rpi 4 ?

Mon Jul 08, 2019 4:36 pm

jahboater wrote:
Mon Jul 08, 2019 4:29 pm
ejolson wrote:
Mon Jul 08, 2019 3:16 pm
The calculation which illustrated the difficulties with 3V power on the Pi 3B most dramatically, in my opinion, was the high-performance Linpack available here. The reported speed of the 4B is about 10.9 Gflops using the power-over-Ethernet hat as the cooling solution. While that number seems reasonable, an independent confirmation would be nice.
I seem to remember in the past, there was a link to a pre-built executable?

The build fails and I would rather be watching Wimbledon that trying to correct the header file errors!
Wimbledon sounds even better than pulling weeds.

I think you need to compile OpenBLAS first and then edit the Makefile for HPL to fix all the paths by hand. The previous binary was created for the Cortex-A53. It is likely a correctly configured A72 version of OpenBLAS will yield a different result as this is one of those widely used libraries with lots of hand-tuned assembler.

User avatar
Imperf3kt
Posts: 2544
Joined: Tue Jun 20, 2017 12:16 am
Location: Australia

Re: Heatsinks official for rpi 4 ?

Mon Jul 08, 2019 10:40 pm

piglet wrote:
Mon Jul 08, 2019 3:01 pm
RichardJones wrote:
Mon Jul 08, 2019 11:13 am
If size isn't an issue but absolute performance is then this active solution would take some beating..
https://www.seeedstudio.com/ICE-Tower-C ... -4097.html
I think it's safe to say this isn't compatible with the official Pi 4 case!
What's holding that on!?
From other posts I've seen, it comes with a metal bracket that screws into the mounting holes on a Pi.


For users claiming the Pi3 cannot run CPU burn A53, I implore you try again. It runs quite comfortably on my Pi3b and still runs fine when overclocked to 1250MHz (after which my Pi becomes unstable and crashes) with no undervolt or throttling.
55:55:44:44:4C
52:4C:52:42:41

ejolson
Posts: 3231
Joined: Tue Mar 18, 2014 11:47 am

Re: Heatsinks official for rpi 4 ?

Mon Jul 08, 2019 11:02 pm

Imperf3kt wrote:
Mon Jul 08, 2019 10:40 pm
piglet wrote:
Mon Jul 08, 2019 3:01 pm
RichardJones wrote:
Mon Jul 08, 2019 11:13 am
If size isn't an issue but absolute performance is then this active solution would take some beating..
https://www.seeedstudio.com/ICE-Tower-C ... -4097.html
I think it's safe to say this isn't compatible with the official Pi 4 case!
What's holding that on!?
From other posts I've seen, it comes with a metal bracket that screws into the mounting holes on a Pi.


For users claiming the Pi3 cannot run CPU burn A53, I implore you try again. It runs quite comfortably on my Pi3b and still runs fine when overclocked to 1250MHz (after which my Pi becomes unstable and crashes) with no undervolt or throttling.
As already mentioned, the problem using cpuburn-a53 as a stress test is that it does not check for errors. It was only designed to test heatsinks by producing heat.

User avatar
Imperf3kt
Posts: 2544
Joined: Tue Jun 20, 2017 12:16 am
Location: Australia

Re: Heatsinks official for rpi 4 ?

Tue Jul 09, 2019 12:24 am

I was implying that it doesn't (anymore) make a Pi3b crash, as had been stated.

I wasn't discussing it's accuracy
55:55:44:44:4C
52:4C:52:42:41

W. H. Heydt
Posts: 10585
Joined: Fri Mar 09, 2012 7:36 pm
Location: Vallejo, CA (US)

Re: Heatsinks official for rpi 4 ?

Tue Jul 09, 2019 12:45 am

RichardJones wrote:
Mon Jul 08, 2019 11:13 am
If size isn't an issue but absolute performance is then this active solution would take some beating..
https://www.seeedstudio.com/ICE-Tower-C ... -4097.html
I think it's safe to say this isn't compatible with the official Pi 4 case!
That one has the advantage of being properly anchored and one could use good thermal compound like, say. Arctic Silver.

jahboater
Posts: 4595
Joined: Wed Feb 04, 2015 6:38 pm

Re: Heatsinks official for rpi 4 ?

Tue Jul 09, 2019 1:54 am

Imperf3kt wrote:
Tue Jul 09, 2019 12:24 am
I was implying that it doesn't (anymore) make a Pi3b crash, as had been stated.
That's correct. In the early days of the Pi3, cpuburn would make it crash in less than 10 seconds.
The engineers did a great job over time in fixing the voltages and it became reasonably stable.

The 3B+ went further by increasing the current available to the cores and dramatically improving the thermal management.

mkirk
Posts: 30
Joined: Tue Mar 06, 2018 4:21 pm

Re: Heatsinks official for rpi 4 ?

Tue Aug 20, 2019 9:10 pm

How does power consumption and CPU temperature vary with core loading (e.g. sysbench, stress or cpuburn-a53) ?

Do these generate same amount of heat and power consumption?
A) 1 core at 100%, 3 cores idle at max clock rate
B) 2 cores at 100%, 2 cores idle at max clock rate
C) 4 cores at 100% at max clock rate

Am I correct to understand that all cores run at max clock rate (e.g. 1.2GHz, 1.4 GHz), even if only 1 core is 100% loaded?
Min clock rate (e.g. 600 MHz) is only seen with all 4 cores idle?

Regards,
Mike

Return to “General discussion”