some questions regard the H3 boards

bronco

hojnikb replied at 2015-11-23 23:06
I think it would make more sense to have default configuration for up to 1200Mhz, since everything ...

I still believe this is not the most important problem (the clock speed). Since I still didn't receive my Orange Pis and none of the people that might easily verify this gives it a try here these are still assumptions (based on fex files, a script.bin I read out and looking into loboris' github repos)

Based on posts here in the forums it's obvious that the SoC even gets way too hot when being idle (~40°C above ambient is insane!). Looks like this is caused by

- only 2 dvfs table entries on the upper limit: 1.53GHz @ 1.5V and 1.2GHz @ 1.3V (according to comments in Allwinner's SDK max. voltage should be 1.4V and max. cpufreq 1.2GHz)
- memory controller clocked to 672 MHz (in Allwinner's SDK 576 MHz were used instead and I still ask myself whether that's not 288 in reality -- but that's a different story)

And then it seems a cpufreq governor is default that switches between scaling_max_freq and scaling_min_freq based on load. But if memory overclocking and overvolting OPPs in the dvfs table are the real origin of the thermal problems than this won't help that much. And at this point I've to wait until my own boards arrive since none of the experienced users wants to try that out.

It seems the H3 based Orange Pis are also wasting enormeous amounts of energy when being totally idle. Which is just insane. It's one thing to provide extra peak performance wasting extra peak energy but normally the CPU cores are most of the time idle and then they shouldn't waste energy. And normally with Cortex-A7 it should NOT make that much difference whether the CPU is idling at 1.2 GHz or 120 MHz: the SoC should stay cool. With the H3 settings used here it seems it stays hot instead... and then there's something wrong with other settings.

You can help of course if you provide me with the output of the following commands to get a clue how default settings look like:

1. cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_governors
   
2. cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
   
3. 
   
4. cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
   
5. cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
   
6. cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq
   
7. cat /sys/devices/system/cpu/cpu0/cpufreq/stats/*
   
8. 
   
9. cat /sys/devices/platform/sunxi-ddrfreq/devfreq/sunxi-ddrfreq/cur_freq
   
10. cat /sys/devices/platform/sunxi-ddrfreq/devfreq/sunxi-ddrfreq/min_freq
    
11. cat /sys/devices/platform/sunxi-ddrfreq/devfreq/sunxi-ddrfreq/max_freq
    
12. 
    
13. ls -laR /sys/devices/platform/sunxi-ddrfreq/devfreq/
    

Copy code

But it doesn't matter if I've to wait. Just realized that loboris built his kernels with 'CONFIG_DEBUG_FS=y' so I'm really curious what's to discover when diving into /sys/kernel/debug
Most important information so far for me: After loboris provided me with some informations I'm certain the heat/consumption problems can be solved easily for my use cases (letting OPi PCs do some work in the most energy efficient way possible)

bronco

And the output of the following would also be interesting:

1. dmesg | egrep "ddrfreq|dsm|cpu_freq"

Copy code

makama80

I see a lot of questions regarding temperature. My orange pi + (H3 board) runs about for 3 months on an interactive governor for 3 months now.

Max. freq. is 1536 MHz, min I have set to 240 MHz. I do have a small aluminium heatsink attached + a small 5v fan. I see rarely temperatures of 80 degrees C on full load. Using the stress tool on full load it's approx. 75 degrees C.

When idle, it's about 40 - 45 degrees. So nothing to worry about.

bronco

makama80 replied at 2015-11-26 02:53
I see a lot of questions regarding temperature. My orange pi + (H3 board) runs about for 3 months on ...

Are you able to execute the following command as root and post the output here?

1. dmesg | egrep "ddrfreq|dsm|cpu_freq"

Copy code

Thx

makama80

Sure... I shortened it a little bit, but values are more ore less the same...

[162320.710078] [ddrfreq] temperature=62 C, ddr freq up
[162321.220085] [ddrfreq] temperature=58 C, ddr freq up
[162327.220126] [ddrfreq] temperature=61 C, ddr freq up
[162327.730083] [ddrfreq] temperature=66 C, ddr freq up
[162329.730073] [ddrfreq] temperature=66 C, ddr freq up
[162330.240105] [ddrfreq] temperature=66 C, ddr freq up
[162331.740074] [ddrfreq] temperature=65 C, ddr freq up
[162332.750087] [ddrfreq] temperature=65 C, ddr freq up
[162334.750116] [ddrfreq] temperature=66 C, ddr freq u

bronco

makama80 replied at 2015-11-26 05:02
Sure... I shortened it a little bit, but values are more ore less the same...

Thx, I forgot to mention that dmesg output is only interesting shortly after a reboot.

bronco

makama80 replied at 2015-11-26 02:53

I see rarely temperatures of 80 degrees C on full load. Using the stress tool on full load it's approx. 75 degrees C. When idle, it's about 40 - 45 degrees

Same here. No heatsink attached, no fan used. No overclocking, no overvolting. Just sane dvfs settings.

This is the OPi PC that arrived an hour ago running "sysbench --test=cpu --cpu-max-prime=20000 run --num-threads=4":

bronco

And this is how it looks if you're not following the overvoltage route but go instead in the other direction. I did some undervolting, used cpuburn-a7 and cpufreq-ljt-stress-test to ensure stability and data integrity and gave sysbench again a try:

Exactly the same performance @ 1.2 GHz (still 158.2 seconds) but 10°C less under full load caused by 80mV less voltage:

faddat

hojnikb replied at 2015-11-23 23:06
I think it would make more sense to have default configuration for up to 1200Mhz, since everything ...

I think that OrangePi has been very careful *not* to falsely advertise their goods.  They're just downstream of allwinner, let's remember.  

Let's also remember:  

Steven figured out how to build and sell a $15 fully-functioning computer.  

While there's significant grumbling about the kernel, and stuff like that, I can't really blame oPi/steven.  They did something very difficult to hit that price point, and should be highly praised for doing so.  

bronco

faddat replied at 2015-11-27 15:19
I think that OrangePi has been very careful *not* to falsely advertise their goods.  They're just  ...

Do you still believe that if you have in mind that only Xunlong shipped their devices with insane overclocking/overvolting settings and that any other H3 device is marketed as and runs with "up to 1.2 GHz"?

Please compare with:
http://linux-sunxi.org/Xunlong_Orange_Pi_2#Manual_build
http://linux-sunxi.org/Orange_Pi_PC#CPU_clock_speed_limit
http://www.cnx-software.com/cate ... allwinner-h-series/

The H3's CPU performance/clockspeed ratio is rather bad compared even to other Allwinner SoCs. So maybe they thought that would be the only way that this slow SoC can shine: By overclocking it from the very beginning leading to all sorts of thermal and stability troubles the forum here is full of.

Regarding the price tag. The H3 is a perfect SoC to build cheap systems (without that much other 'active' components on the PCB). Needs no PMU/PMIC, no external Ethernet PHY due to the integrated (100 Mb/s only) and also contains 4 (!) USB-PHYs already. And please remember that this approach (combining a SoC with just some passive components around) is also possible for 1/3 of the price (the hardware of the RPi Zero and OPi PC aren't comparable at all, it's about the design principle to use everything the SoC already provides and to not rely on additional 'expensive' components on the PCB: something that started just with the Orange Pi PC here).