NanoPi M4 and overheating running BOINC

[seven.7777777s]

I am testing a Nano-Pi M4 with 4GB of RAM, with the heat sink and fans installed

ARMBIAN 5.65 stable Ubuntu 18.04.1 LTS 4.4.162-rk3399, docker CE and BOINC / seti@home, installed.

I am using 7 Raspberry-Pi 3B+ boards as nodes.   I would like to implement and test boinc-client-docker, (https://github.com/BOINC/boinc-client-docker).

I am new at this and running into overheating issues, (temps over 70C as reported by htop), when all 6 cores are maxed out at 100%.

I can keep the board slightly below 70C if I add multiple fans pointed at the heat sink, however I would like a better solution.

 

I was wondering if setting the max frequency of the cores to a slightly lower level would resolve this issue.

2x Cortex-A72 at up to 2.0GHz, 4x Cortex-A53 @ up to 1.5GHz

cores 1 through 4 will run at 1.49x and cores 5 & 6 will run at 1.99x

they all idle at 600 MHz with a resting temp of 32C

Would setting thee max frequencies to 1.48 and 1.8 help resolve this issue?

What would be the best method of testing this?

 

The results of

cat /etc/default/cpufrequtils
# WARNING: this file will be replaced on board support package (linux-root-...) upgrade
ENABLE=true
MIN_SPEED=600000
MAX_SPEED=2016000
GOVERNOR=ondemand

 

Edited November 1, 2018 by seven.7777777s

[chwe]

Indeed it would. Less voltage on the CPU normally ends in less heat.. CPU supports dvfs so by lowering the max CPU freq normally helps. But why is the 70°C an issue? E.g. the CPU throttles if it's to warm. Just your expectation that it is to warm or do you have issues with it?

 

https://github.com/armbian/build/blob/master/patch/kernel/rk3399-default/overclock-rk3399-to-1.5-2.0.patch

is responsible for the 1.5 GHz/2GHz frequency... Not sure if adjusting 'cat /etc/default/cpufrequtils' also affects little cores (not that familiar big.LITTLE). Otherwise you would have to adjust the DT. And maybe apt-hold the related package to ensure it doesn't get overwritten by armbian updates.

 

might be worth to read.

[seven.7777777s]

Without extra cooling fans the Nano-Pi M4 board is exceeding 75C and shutting itself down.

Thanks for the links to articles,  I will read them.

[NicoD]

37 minutes ago, seven.7777777s said:

I was wondering if setting the max frequency of the cores to a slightly lower level would resolve this issue.

2x Cortex-A72 at up to 2.0GHz, 4x Cortex-A53 @ up to 1.5GHz

 

Those are the normal temperatures. With the Lubuntu from FriendlyElec the temperature is a lot lower since it`s clocked to 1.4Ghz/1.8Ghz.
 

Temperature
-----------
Armbian Bionic/Stretch 64-bit 2Ghz + 1.5Ghz
						With fan idle    32°C
						With fan maxed   65°C
						No fan idle      40°C
						No fan maxed     Throttles at 85°C after 14m30s
						
Lubuntu armhf/arm64 1.8Ghz + 1.4GHz
						With fan idle    29°C
						With fan maxed   54°C
						No fan idle      42°C
						No fan maxed     69°C (after 30 minutes maxed)

The fan works better when it blows over a big area by hanging crooked. You also have to raise the heatsink with screws so the airflow can pass. Just don`t let is sit at +80C.
You can downclock with cpufreq.

[NicoD]

2 minutes ago, seven.7777777s said:

Without extra cooling fans the Nano-Pi M4 board is exceeding 75C and shutting itself down.

Thanks for the links to articles,  I will read them.

It`s not normal that it shuts down at 75C. This isn`t a too high temp. It should throttle at 85C. But I wouldn`t let my board run long at such temp.

[seven.7777777s]

Running the Lubuntu supplied by the board vendor I was able to run BOINC for over 24 hours and the temp with a single fan was about 60C; however I had no success installing Docker CE.

 

Running Armbian With a single fan running BOINC the board is shutting down after about 20 minutes.  I was able to successfully install Docker CE and attach the 7 Raspberry Pis as nodes..

 

How would you recommend resolving this?

 

[NicoD]

Check your voltage on the usb3 port. Maybe you`ve got an undervoltage.
It should not shut down at these temps. I`ve done a blender benchmark for +1h at 85C. No problem.
Maybe because one side has no air flow it heats up a lot. Miners also heat up more than other applications. No idea why. Best to underclock if you`re mining.

This is how I run it. Maxed out in Bionic it reaches 70C with this fan.

[seven.7777777s]

Thanks, I will check voltage, as that looks like the most likely cause.

 

[seven.7777777s]

I received a Drok Model UM25 tester, and started checking the voltage and amperage used.  I noticed that the NanoPi M4 when running at between 98% and 100% on all 6 cores draws a little over 5.1V and between 1.9A to 2.2A,  This is getting close to the maximum of 5A and 2.5A that the cable, (Anker Powerline+ II USB-C to USB-A 2.0 Cable (3ft)), and power supply, (Anker 60W 10-Port USB Wall Charger), are rated.

I shut down the 7 Raspberry PI 3B + devices also on the power supply and watched the NanoPi M4 use over 5.3V and 2.59A then reboot.

I looked up the specs for the NanoPi M4, it needs 5V 3A

The RaspberryPi 3B+ needs 5V 2.5A, (which is where the power supply and cables I am using are rated)

I ordered a 5V 4A power supply and a cable rated for 4A.

Until they arrive I an running the NanoPi M4 at 1.42GHz and 1.80GHz using the following commands

cpufreq-set -c0 -u1.42GHz
cpufreq-set -c1 -u1.42GHz
cpufreq-set -c2 -u1.42GHz
cpufreq-set -c3 -u1.42GHz
cpufreq-set -c4 -u1.80GHz
cpufreq-set -c5 -u1.80GHz

It is now using 5.2V between 1.61A to 1.82A and does not reboot and the temp is showing 53C

 

[seven.7777777s]

OK, I obtained a 5v 4A power supply and cables rated for 5V 4A.  I also added supports using , (uxcell 50pcs M3 10+6mm Male Female Thread Nylon Hex Standoff Spacer Screws PCB Pillar White), see attached picture.  The temperatures are in the 53C to 60C range.  The board's electrical usage can burst up to 2.85A @ 5.1V

Running 'Armbian_5.65_Nanopim4_Ubuntu_bionic_default_4.4.162_desktop' on an 32GB or 128GB Samsung microSDXC UHS-I  card.

I was able to successfully install Docker CE and create a swarm, CPU cores are fine un-throttled until I start running BOINC tasks.

A stable environment running BOINC's Seti@home requires me to throttle the CPU at

cpufreq-set -c0 -u1.42GHz
cpufreq-set -c1 -u1.42GHz
cpufreq-set -c2 -u1.42GHz
cpufreq-set -c3 -u1.42GHz
cpufreq-set -c4 -u1.80GHz
cpufreq-set -c5 -u1.80GHz

 

Running 'rk3399-sd-friendlydesktop-bionic-4.4-arm64-20180921' on an 32GB Samsung microSDXC UHS-I  card.

I was unsuccessful in installing Docker CE

A stable environment running BOINC's Seti@home requires me to throttle the CPU at

cpufreq-set -c0 -u1.20GHz
cpufreq-set -c1 -u1.20GHz
cpufreq-set -c2 -u1.20GHz
cpufreq-set -c3 -u1.20GHz
cpufreq-set -c4 -u1.61GHz
cpufreq-set -c5 -u1.61GHz

 

I will be experimenting with some copper heat sink shims and thermal paste to reduce temps.

I am still working on getting seti@home/BOINC to run in Docker across a swarm.  The instructions that are out there will require me to learn more in order to successfully implement them.

[seven.7777777s]

Screen shots from rk3399-sd-friendlydesktop-bionic-4.4-arm64-20180921' on a 32GB Samsung EVO Plus microSDXC UHS-I  card.

running BOINC / seti@home.  This image was updated and upgraded to the latest version available this morning.

At this point I am reasonably certain that the instability I am experiencing with BOINC is due to issues with the manner in which the Rockchip RK3399 CPU has been  implemented in the OS.  The ''Armbian_5.65_Nanopim4_Ubuntu_bionic_default_4.4.162_desktop being more stable under heavy load than the ''rk3399-sd-friendlydesktop-bionic-4.4-arm64-20180921.

I ameliorated the known power and cooling issues; the remaining variable is the OS build.

[NicoD]

@seven.7777777s

Nice work. Replacing that thermal pad with copper shims will do a lot.
Also see that your RPi 3b+`s don`t go over 60C. It clocks to 1.2Ghz above 60C. The governor will still say it`s at 1.4Ghz.
You could use 5v fans on your pi`s instead of large fans blowing. Most energy from a large fan further away goes to waste. The more the airflow is aimed at the heatsink, the better.

What`s the load you`re giving this? Is it for mining or so? Good luck.

[seven.7777777s]

Thanks for the tips  The load is from BOINC with tasks from seti@home

Installing BOINC on Ubuntu

https://boinc.berkeley.edu/wiki/Installing_BOINC_on_Ubuntu

 

[seven.7777777s]

Here's a screen shot of the load presented to Armbian_5.65_Nanopim4_Ubuntu_bionic_default_4.4.162_desktop where I can run the cores at a higher level than on friendlydesktop-bionic-4.4-arm64-20180921. 

[dogshome]

On 11/8/2018 at 12:21 AM, NicoD said:

@seven.7777777s

Nice work. Replacing that thermal pad with copper shims will do a lot.
Also see that your RPi 3b+`s don`t go over 60C. It clocks to 1.2Ghz above 60C. The governor will still say it`s at 1.4Ghz.
You could use 5v fans on your pi`s instead of large fans blowing. Most energy from a large fan further away goes to waste. The more the airflow is aimed at the heatsink, the better.

What`s the load you`re giving this? Is it for mining or so? Good luck.

Copper shims make about 15C difference :-)

[seven.7777777s]

@dogshome

Thanks for the tips,

The load is from BOINC with tasks from seti@home

https://setiathome.berkeley.edu/

Installing BOINC on Ubuntu

https://boinc.berkeley.edu/wiki/Installing_BOINC_on_Ubuntu

[shaun27]

On 11/9/2018 at 3:57 PM, dogshome said:

Copper shims make about 15C difference :-)

Agree with that.  over 22c difference for me.

 

Anyone wants to know what size shims i used 15mm by 15mm and 1.2mm thickness with a dab of thermal paste on cpu and topside of shim to main heatsink.

 

Temperature difference

 

Exactly same test with youtube 1080p 60c with thermal pad, 38 to 40c with copper shim and thermal paste each side of it.

 

 

[seven.7777777s]

Thanks for the feedback, I ordered shims, I will post the results when they arrive

[seven.7777777s]

I received shims and added a 20 x 20 with 1.2 thickness and thermal paste, to each NanoPi M4 the temperature is now at 45c with a full load.

I edited /etc/default/cpufrequtils & changed the max speed setting and rebooted to keep the two big cores at 1.80GHz

MAX_SPEED=1800000

Then rebooted

Dual fans attached to the heat sink help.  I am using

ANVISION 2-Pack 40 x 40 x 10mm 4010 Dual Ball Bearing DC 5V USB Brushless Cooling Fan UL CE YDM4010B05

https://www.amazon.com/gp/product/B074YJH3P9/ref=oh_aui_detailpage_o00_s00?ie=UTF8&psc=1

& a battery backup for the RTC

LeFix Replacement CMOS RTC Battery with Cable(2 Pin 2Wire) for HP Compaq G50 G56 G60 G62 G70 G72 CQ40 CQ50 CQ56 CQ60 G62 CQ62 G56 CQ56 CQ72

https://www.amazon.com/gp/product/B07DCP26N3/ref=oh_aui_detailpage_o00_s00?ie=UTF8&psc=1

 

Standoffs to raise the fans and heat sink can be created using 4 x M3 Hex Socket Head Cap Screws in 25MM to 50MM length, search on Amazon for these is below

https://www.amazon.com/s/ref=nb_sb_noss?url=search-alias%3Delectronics&field-keywords=M3+Hex+Socket+Head+Cap+Screws+in+35MM+length

 

The fans are attached to the heat sink using M3 x 12 mm Zinc-Plated Phillips Pan-Head Machine Screw, these are available at Home Depot as well as Amazon

https://www.homedepot.com/p/Everbilt-M3-x-12-mm-Zinc-Plated-Phillips-Pan-Head-Machine-Screw-3-per-Bag-802718/204282659

 

 

[NicoD]

4 hours ago, seven.7777777s said:

received shims and added a 20 x 20 with 1.2 thickness and thermal paste, to each NanoPi M4 the temperature is now at 45c with a full load.

I edited /etc/default/cpufrequtils & changed the max speed setting and rebooted to keep the two big cores at 1.80GHz

Great job. Now that's a "cool" SBC.  That's 10°C less than mine at 1.8Ghz.
Can't you run at 2Ghz now? Mine is 54°C maxed on 1.8Ghz with fan, no copper shim. At 2Ghz it's 65°C. So I expect you would be able to run it at 1.5Ghz/2Ghz at around 55°C. A very acceptable temperature.
 

[seven.7777777s]

 I tried setting the MAX SPEED @ 1992000, it crashes after about 4 minutes with the temp @ 43 C

cat /etc/default/cpufrequtils
# WARNING: this file will be replaced on board support package (linux-root-...) upgrade
ENABLE=true
MIN_SPEED=408000
MAX_SPEED=1992000
GOVERNOR=ondemand

 

I am running 2 x 2gb & 2 x 4GB boards with 5V 4A power supplies with copper shimmed heat sinks and they will run for days at 1800000

This is not a heat issue, it's a stability issue running BOINC at that speed with this OS

 

 

[NicoD]

29 minutes ago, seven.7777777s said:

I am running 2 x 2gb & 2 x 4GB boards with 5V 4A power supplies with copper shimmed heat sinks and they will run for days at 1800000

This is not a heat issue, it's a stability issue running BOINC at that speed with this OS

Is it with every board the same behaveriour? Sorry if I ask again, you did try another short power cable? A 4A power supply is worth nothing when using a bad cable. When running at 1.5Ghz/2Ghz it uses 0.5A more (2A vs 1.5A). So a bad cable could be affected a lot more.
Do you have logs?

What is the crash? Does it reboot? Does it hang? Does it still show anything on HDMI? Is it only when running BOINC? Is it only BOINC that crashes?

All I can say is I never had a crash at 1.5Ghz/2Ghz with Armbian Bionic. Even not when it was underpowered.
I hope you'll find it, I bed you're going to have a lot more performance at 1.5Ghz/2Ghz.

[seven.7777777s]

The same on all 4 boards, I am using cables rated at 5V 4A, two different brands and the cable that came with the supply.  When I check the Amperage going into the cables it peaks at 2.85 A 5.1V from te supplies.  One supply was purchased at Amazon with the 5V 4A supply, three other 5V 4A supplies were purchased from China.  They are the same power supply, just the ones purchased from China are labeled in Chinese characters.

[NicoD]

And is it only with BOINC that it crashes? What kind of crash?
Try another heavy load and see if it also does it.

[seven.7777777s]

What test bench program, would you recommend, that will will place all 6 processor cores at or close to 100% for over 10 minutes?

[jtremblant]

I was really worried my temps were reaching 77 degrees at full load for my new RockPi4C running latest Armbian w/ 5.8.13-rockchip64 kernel. Disappointing seeing Radxa not been able to offer a better heatsink/cooling solution for their SBC's, even if it's more expensive. I ended spending way more that trying to lower my CPU temps. Just ordered a ecoPI PRO HP Aluminum housing for ROCK PI 4C case. Will see how it goes with it.

Thank's to this post, I ordered from amazon some Artic MX-4 thermal paste, Copper shims and a M3 Hex Brass Standoff Male-Female Bolt kit. I managed to lower my CPU temps to an acceptable 57 degrees at full load. That's some impressive 20 degrees less. My Average CPU temps are now around 50 most of the time when I working on my SBC. I got 20mmx20mm and 1.2mm thickness copper shims but seems like I applying some pressure on the PCB corners.

Maybe I should've ordered some 0.8mm thick copper shims instead.

Having Armbian 20.08.7 Focal running at 2.02 and 1.51Ghz on governor performance is a nice feature.