OrangePi Plus 2 and OrangePi 3 are coming!

makama80

Eh... Mr. Bronco... that's not entirely true. There's quite some H3 and A83T activity on linux-sunxi.org and actually some people managed to boot a 4.x kernel on H3 boards recently. Check the IRC log on linux-sunxi.org and this forum. You will be surprised what you will find!

That Allwinner dropped support does not really make any significant difference: there has never been any good support at all from 'Big A'. Allwinner is a chip maker, not a software maker.

Looking at the A20 for example: that's even an older chip and still going strong in the community. Currently the H3 is a rising star. Last week some gurus from this forum even managed to awake the Mali acceleration in Linux...

And I dare to challenge if you really need a Linux 4.x kernel  or Android 5.x for your application(s). People always scream that they want the newest software. What's wrong with proven technology as long as it suits your needs?

bronco

makama80 replied at 2015-11-20 17:45
Eh... Mr. Bronco... that's not entirely true. There's quite some H3 and A83T activity on linux-sunxi ...

I monitor all these activities very closely. In fact Jens is doing some work to get mainline u-boot/kernel support for H3 and Vishnu made some first steps with A83T.

Both SoCs are targeted at clueless people that believe the more cpu cores you get the faster everything will be (same story with 32 vs. 64 bit and "up to" cpufrequency marketing chitchat). These seem to be the marketing requirements to sell Android tablets and OTT boxes and that's what these SoCs are made for. You can't sell a device when it's labeled dual-core even if its faster than its quad-core successor.

Both SoCs don't get any support from Allwinner any more (in the form of an SDK). Therefore the H3 will only be useable with Android 4.4 (hardware accelerated video decoding) and the A83T receives support for Android 5.1. Running both SoCs with Linux isn't interesting at all in my opinion except Orange Pis being dirt-cheap.

The forums here are full of heat problems and stability issues. Seems like all this is caused by insanely overclocked default settings. If the H3 is operated with sane values then both problems are gone but its integer performance is a bit below RPi 2. But on the latter you can use HW accelerated video which will not happen with the H3 (and yes, Steven will say "give us 2-3 weeks" as he does every few weeks and people will believe him again and again).

I'm happy to learn recently that the H3 is not broken by design as it seemed prior to Olimex getting their hands on their first H3 board. It took them only 24 hours to realize where the problems originate: In the fex files used here. I just ordered one dirt-cheap Orange Pi PC to play with the crappy 2MP camera module and evaluate it for home automation and GPIO stuff (after clocking it down to the minimum und deactivating 2-3 CPU cores). That's the only use case I see where H3 based Orange Pis match

One final word regarding Banana Pi M3: software situation even worse than with the H3, zero support by the manufacturer (SinoVoip doesn't even get the idea what support would be), slow single-core GPU and as usual no HW acceleration within Linux, slow pseudo-SATA implementation and most likely also low single-threaded performance. But for people that think 'the more cores the better' it would be the perfect fit.

bronco

makama80 replied at 2015-11-20 17:45
And I dare to challenge if you really need a Linux 4.x kernel  or Android 5.x for your application(s)

Regarding this: Some of our use cases for ARM boards do really benefit from mainline kernel support. I've already written two articles in the linux-sunxi wiki trying to cover this:

http://linux-sunxi.org/Sunxi_devices_as_NAS

http://linux-sunxi.org/USB/UAS

To sum it up: If you plan to use a sunxi device as a NAS performance with mainline kernel will always be superiour. If you have to use USB 2.0 with mainline kernel you're able to use UAS which really helps with performance (but you've to choose a good USB-to-SATA bridge and this does neither apply to the Orange Pi Plus nor the Banana Pi M3 since they both use an outdated crappy GL830 chip that is horribly slow and not UAS compliant). When using USB 2.0 depending on the bridge chip you use it might be a difference by factor 2 being limited to the old sunxi 3.4 kernel or being able to use mainline. And only with mainline kernel you're able to use modern file systems like btrfs which helps both with performance (due to transparent filesystem compression increasing possible throughput) as well as storage efficiency and data integrity on top. But you can't use btrfs with sunxi's 3.4 kernel.

Fun fact: If you plan to use an SBC as a NAS then every A20 based device featuring both SATA and native GBit Ethernet will easily outperform any H3 or A83T based. Because the amount of available CPU cores doesn't matter at all. So the 'slow' A20 with its mature mainline kernel support and its unique hardware features compared to H3/A83T is the best choice. And since we now know that Allwinner plans to relase a pin-compatible A20 successor in the second half of 2016 with 4 A7 cores I feel no need to even have a look at something called Banana Pi M3.

geev03

>So the 'slow' A20 with its mature mainline kernel support and its unique hardware features compared to H3/A83T is the best choice
I used to stress test an A10  ( by using it as a  moodle 2 server. A20 was a bit better.
Compared to those ,the H3 Orange PiPC is rock steady as a debian server- serving moodle 3 etc. A similar Raspberry Pi2 server could hardly stay on for a couple of days before rebooting was needed...
The H3 server is going strong even after three weeks.

jacer

geev03 replied at 2015-11-23 01:34
>So the 'slow' A20 with its mature mainline kernel support and its unique hardware features compared ...

Now we only need mainline kernel support and HW playback.

igorpec

http://www.amazon.com/pcDuino-pcDuino3-Nano-Lite/dp/B00ZEPZGQO
A20 device for 15USD

It doesn't have 4 cores but It's better (almost) in every way but heating the room ;)

bronco

igorpec replied at 2015-11-23 13:27
http://www.amazon.com/pcDuino-pcDuino3-Nano-Lite/dp/B00ZEPZGQO
A20 device for 15USD

Thx for the link, just realized that they also sell the A10 based pcDuino2 for 15USD (Wi-Fi and NAND included, no GBit Ethernet as expected but also no SATA). I always searched for a cheap A10 board... but with SATA and more for historical reasons ;)

Regarding heating the room: I'm still convinced that the heat problems H3 users suffer from can be resolved easily Yesterday I decided to start with these settings when my OPi PCs arrive:

1. boot_clock      = 1008
   
2. dram_clk        = 480
   
3. 
   
4. [dvfs_table]
   
5. pmuic_type = 2
   
6. extremity_freq  = 1536000000
   
7. max_freq        = 1200000000
   
8. min_freq        = 108000000
   
9. LV_count        = 8
   
10. LV1_freq        = 1200000000
    
11. LV1_volt        = 1320
    
12. LV2_freq        = 1104000000
    
13. LV2_volt        = 1250
    
14. LV3_freq        = 1008000000
    
15. LV3_volt        = 1200
    
16. LV4_freq        = 960000000
    
17. LV4_volt        = 1160
    
18. LV5_freq        = 816000000
    
19. LV5_volt        = 1100
    
20. LV6_freq        = 540000000
    
21. LV6_volt        = 1040
    
22. LV7_freq        = 300000000
    
23. LV7_volt        = 960
    
24. LV8_freq        = 108000000
    
25. LV8_volt        = 960

Copy code

I already finished a first script that will iterate through these settings and lower voltage by 20mV for every cpufreq available, adjust script.bin, reboots and let cpuburn-a7/cpufreq-ljt-stress-test check the stability until the first error occurs. When this is done, I repeat the whole thing one time with "dram_clk = 360" and CPU cores disabled and another time again with all 4 cores and "dram_clk = 576". Then the voltage settings will be increased a little to be on the safe side, then lima-memtester has to confirm data integrity and then it's time for benchmarking

I hope my OPi PCs arrive soon.

The H3 contains the same old boring Cortex-A7 cores Allwinner uses since 3 years in all low-end SoCs but made in a more efficient process (28nm vs. 40nm). Why should this chip overheat unless it's overclocked in an insane way (heating up like crazy even when idle -- this is something that puzzles me)

igorpec

https://olimex.wordpress.com/201 ... ototypes-are-ready/

I guess heat is mostly due to onchip NIC ...

bronco

igorpec replied at 2015-11-23 15:59
I guess heat is mostly due to onchip NIC ...

Wrong first assumption. One day later they tested their new A33 board (sharing the same amount of old boring Cortex-A7 cores with the H3 but in the more inefficient 40nm process) and there everything was ok, then they had a look into H3 again and found the culprit: https://olimex.wordpress.com/201 ... test/#comment-20502

As expected. Xunlong started to advertise falsely the H3 being able to run at 1.6GHz back in April (but knew already at the same time that any cpufreq above 1.2 is for benchmarking only), the fex files used here are insanely overclocked and the result is that anyone thinks the H3 has to be as hot as possible and needs heatsink and fan to operate reliable even when idle. I still can't believe it

igorpec

OK, I overlooked this ... crazy overclocking.