BPI-M3 Heatsink


Every post from charles is adversary troll

Would you have such a ‘Cooler’ for me too? :wink:

It looks perfect. Is it alu?

Greetings from Austria

They are expensive for me to make +200 USD I can if you have the budget. My lab is not set up for production only prototyping so it is very costly for small items.

I am forwarding the design to Sinovoip to find asian casting at much lower cost.

If you have access to a 3d printer you may want to form a bracket that will hold a copper northbridge cooler. A cooler like this one will work nicely as well. http://www.acousticpc.com/swiftech_mcx159cu_all_cooper_universal_north_bridge_chipset_cooler.html

I posted REAL benchmarks and Results.

Only +200 USD, that’s a bargain! Especially for the ‘performance gain’!

Ok, you got with your super cheap heatsink sysbench 20000 results of less than 52 seconds or with the other test a little above 53 seconds (what’s that ondemand/performance thing?). Now let’s google for ‘sysbench 20000 banana pi m3’:

  • first hit: ‘$0.5 heatsink that performs ok’ and ‘sysbench --test=cpu --cpu-max-prime=20000 run --num-threads=8" in less than 53 seconds’
  • second hit: the same guy asking @sinovoip but to no avail
  • third hit: without heatsink it’s 92 seconds

Impressive results! For just $200 REAL performance improves by unbelievable 75 percent compared to no heatsink. And when comparing to a heatsink that costs 400 times less it’s at least an improvement of ZERO percent. REAL benchmarks and not FAKED ones as this cpuminer stuff above!

Good luck!

Let’s see your numbers with my provided software.

You are proving my point for me…

I am not currently selling the heat sinks as you can see. I am providing the design to Sinovoip Inc. threw Ammorexue. The goal is to get them made inexpensively in asia.

You’re so funny. You provide REAL benchmarks using sysbench 20000, get a result of ~52 seconds and don’t understand the meaning of other people without your AMAZING or let’s say REAL heatsink scoring 53 seconds with just an average and cheap heatsink?

Good luck!

minerd is not a benchmark it is a test for bit coin mining that is one calculation iterated.

I have provided you and all with a tool that allows you to verify your work.

I have verified my work and my needs nothing you can do in your world will change the fact that I can successfully use the BPI-M3 for my needs and you have not.

Really? :joy::joy::joy:

I showed you already my khash/s values with my cheap heatsink so that you’re able to understand why your milled heatsink is just an insanely expensive overkill. It’s so easy for you to verify that your heatsink is not worth the efforts. All that’s needed is you starting to understand what’s happening around. Good luck!

You should provide the logs instead of just refuting and trolling.

Load levels on the none standard tests hit 72 not 8 like minerd .

Every post you have ever made on this site provides no benefit only refutation.

Dear auto,

here you find the results as requested: BPI-M3 Heatsink (choose the latest directory called 09-19-46 there)

Not so surprisingly the numbers I got are identical to yours since I made two important improvements in the meantime:

  • BGAs dissipate heat via contacts to the PCB. Allowing some airflow below the PCB improves heat dissipation.
  • Improving the airflow (lateral) on top of the heatsink is more important (of course only as long as we’re talking about improving heat dissipation and not your ‘My heatsink is longer than yours’ game)

You should improve your scripting skills a bit, having to run all of this as root is bad style, not checking for external requirements (‘parallel’ command and the others) is bad style too and your error handling is also bad (ramdisk creation failed). In the end I had to start multiple times your time consuming and mostly useless ‘benchmark’ so please check logs/2017/06-21/08-27-27 log directory to draw the graphs you seem to need.

I was running with interactive cpufreq governor (since the only reasonable choice with this kernel) and my sysbench numbers look like this:

total time:                          51.7640s
total time:                          51.7103s
total time:                          51.7564s
total time:                          51.7369s
total time:                          51.7183s

Yours with performance governor:

total time:                          51.7284s
total time:                          51.6777s
total time:                          51.6585s
total time:                          51.6866s
total time:                          51.6727s

Yours with ondemand:

total time:                          51.6665s
total time:                          52.7399s
total time:                          52.4297s
total time:                          53.1433s
total time:                          53.3066s

I fail to understand the purpose of the other ‘benchmark’ (this funny palindrome thingie) since we’re talking about the efficiency of heat dissipation – how do you want to test this only with very light loads as you always do (you seem to be confused about the concept of ‘average load’ anyway: http://lmgtfy.com/?q=average+load+linux). Of course I won’t repeat those stupid and time consuming tests again with performance governor to show you exactly the same numbers as you got.

If you would’ve chosen cpuminer as suggested by me you would’ve long known that your heatsink is just the expensive and also inefficient overkill as it looks like :slight_smile:

BTW: My board still runs powered by Micro USB from a $7 5.1V/2A PSU but with an appropriate cable avoiding voltage drops.

In case you need further help, eg. how to improve thermal behaviour (fex file magic) or to get higher clock speeds (fex file magic reloaded) just drop me a note. But don’t forget to insult me as you do all the time! It seems to be important to you trashtalking to the persons that try to help you. Don’t give up on this!

PS: Another simple way to test heat dissipation efficiency besides cpuminer is using cpuburn-a7 and then watching how low cpufreq drops or even better check /sys/devices/system/cpu/cpu0/cpufreq/stats/time_in_state prior to 2 minutes cpuburn-a7 and after that and do the math. Already curious about your (low) results :slight_smile:

Nice, you did not even check for stress-ng but think this would be important.

That’s how it looks like after installing all your requirements:

Starting Stress-ng 5 minute run 
Temp: 72C----------Load average: 8.10, 8.49, 20.62

Starting Stress-ng 10 minute run 
Temp: 72C----------Load average: 8.17, 8.39, 19.00

It’s just another light load not able to heat the whole thing up near the throttling treshold. Useless :slight_smile:

Ok, another run, this time with performance governor after heating up stupidly for 15 minutes before (if temperature remains stable a few degree below throttling treshold what should happen?):

total time:                          51.5482s
total time:                          51.5281s
total time:                          51.5073s
total time:                          51.5191s
total time:                          51.5412s

These were yours

total time:                          51.7284s
total time:                          51.6777s
total time:                          51.6585s
total time:                          51.6866s
total time:                          51.6727s

(I’m running an Armbian/Jessie userland on this M3, if I would run an Armbian/Xenial userland instead sysbench execution numbers would be as low as 36 seconds. Of course you will neither believe this nor test yourself :joy: )

Dear auto, one last attempt to help you: Since you seem to want to do some heavy stuff on these eight Cortex-A7 cores. Look what’s possible with 8 A53 cores and a heatsink from the bin on a M3 (not Banana but NanoPi this time): https://forum.armbian.com/index.php?/topic/1285-nanopi-m3-cheap-8-core-35/&do=findComment&comment=12872 (also less expensive, able to be powered from a power header by default, manufaturer providing a cheap but efficient fansink and maybe soon running with 64-bit kernel and userland, then sysbench execution times drop below 2 seconds)

You really need to get familiar with the concept of ‘average load’ especially on Linux. Someone such an expert as you should understand what adds to ‘load’ and what not (on Linux, different platforms --> different behaviour, but average load almost everywhere just the wrong metric :slight_smile: )

Here are your logs logs.zip (101,7 KB)

Please think about the next time following my advice to simply use minderd --benchmark since this is already sufficient to get numbers about the efficiency of heat dissipation. :slight_smile:

First Average load and task performance are directly proportioonal on any multi core system. These processors also have a vector unit for each set of 4 cores. You make use of the processor more fully when the load is high. If your load number is low you are not stressing the processor completely.

Many of the single thread benches are attached to single processor instead of available threads. Minerd works this way each runner watches a specific processor for ready before accepting the next task. You will not get a very good utilization doing this. Stress-NG works the same roughly sysbench will make use of the cueing and vector units if you set the process count to 48-64 you will see a trim in the time necessary to do the task. Finally the tasks need to be written to be aggressive themselves an easy way to do that is to loop while a file exists and kill themselves when the file is missing. This way you can launch 64 workers at the same time the workers will also make use of the cache and the look aside in the processor. Parallel allows a simple perl script to launch subshells until the process is done.

CPU-RAM.sh is progressively brutal as it makes the hex code then folds it and searches the cache file before writing and stopping. If entropy is low the task takes forever due to duplicates in the cache file. If entropy is high the script still has to search a variable length up to 10,000 line file.

SO in real world this is the same as doing a pre shared key lookup on an encrypted channel. This is not a public key but two separate keys that are compared on both ends on each packet. Both the client and the server have a random group of pre shared tokens 100 users 100 keys each and packets are keyed with these codes. Each server instance 8 per A83T supports the encryption for 100 users. Or in more practice use 6 cores for encryption 2 cores for other uses.

Charles you are missing three files from the logs. graph.svg Heatsink.log The last few lines of the autosevo report are also missing.

It should have indicated on standard out the watts dissipated by cooling device and the total time for cpu-ram1 and cpu-ram2

I can see in Temperature.csv That you reached peak load at 73 load 2017-06- and began throttling.

At 2017-06- you dropped to 480mhz and remained in overheat at 1.6Ghz and it looks like you lost a couple of cores. Load also dropped to @40

Looking at the Temperature.csv I can see that you throttled and dropped cores and frequency.

Make sure you have gnu plot installed it was part of the standard base LSB if not installed will not create graph. Please install gnu plot and rerun I am curious to see the graph.

And what was the heatsink rating in watts? Curious that your report is missing those lines and the graph.

Also you may want to learn who I am and where I have been. Your comments about what you know to be true are drawn from books I and my friends have written. I started at Penn State in 1968 at the ChiChester campus of the Wistar Institute. They delivered the very first hard drive to our lab in 1971. The original Point to Point over Ethernet was developed by myself and 2 others at Bell Telephone in 1993. I am the author of the Afghan, Pakistan, Punjab telephone system. And I was one of the owners of Frontier Telecom. I created Subka for Mubaric in Egypt and Think Wifi in Pakistan. I have run telephone systems into Cuba, Jamaica, Nicaragua, Punjab, NWFP Pakistan, Kabul Jalalabad Afghanistan. I owned Creative Marketing Impact who was the marketing company for Banana Boat, Pananma Jack, Sea & Ski and 20 others. I designed and marketed children’s toys including Blues Clues, Dora The Explora, Sega Sonic the Hedge Hog , Fischer Price Little People and mr & ms Potato head. In 2007 I built myself number 76 in the top 500 computers. I designed and built the electrical system for the Archistrat RockCity the first computer to run faster than 1Ghz. I have designed wrote and run several bitcoin pools one producing 30 coins per day.

If you ask you may learn if you argue you are nothing more than a troll.

OMG, again so much text no one ever will read. Get the 09-19-46 directory and answer your questions yourself please.


The stress-ng and sysbench runs heat soak what ever heat sink you have then it measures the cool down to calculate specific change in heat energy in joules then calculates the watt dissipation of the heat sink. Thats why it asks for the gram weight of heatsink and the material of heat sink.

You only have to zip the last directory not all of them each time.

IE zip -r zippy.zip /opt/autosevo/logs/2017/06-21/08-27-27

Your cooling method Dissipated : .9242W

The board itself is @4A at full power use at input of 5.7V on header. The board uses 22.8W maxed out. Only cooling .9242W heat will build up and tigger cooling. You need to remove same heat that you are generating for stability.

This is the most basic concept of integrating systems.