RK3328 Kernel

[TonyMac32]

Tinker OS has 1.8 GHz opp point in it's dts, I'll update that.  (Also, the Miqi patches already applied to 4.11 give it 1.8 GHz)

 

The reboot hack/fix on 4.4 does not work on 4.11.  There appears to be an mmc soft-reset property in 4.12, but I haven't tracked down all the parts. 

 

Final question:  does anyone else lose HDMI after an extended inactive period?  I won't discount it being my monitor, but show of hands?

 

Lastly, I picked up https://www.amazon.com/gp/product/B01HSBM5JM/  to replace the 20x20mm tinker board default.  I did not use the included thermal pad, I used a thin 20x20 one I had lying around.  I had to cut a 2x2 fin chunk off to go around the inductor next to the Realtek audio chip.  Now, in my noisy environment I'm not too worried about it, and I was trying to compile some things, so I stuck a little 25x25 mm fan on it directly over the processor.  I was idling around 42 C, now I'm at 32.  Yes, it's a noisy little bugger, but not overly so. (next to my gaming/dev pc it's silent)

 

I have not gotten a real chance to try out running the board from the GPIO +5 and Gnd, however after the system shuts down those pins remain energized, so that's a good sign.

[Tido]

1.8 GHz, I guess it is easy to replace the DTS in the build to keep running it at 1,6GHz. So it is fair enough to change this. On the other hand it would be interesting looking first at different throttling options like reducing frequency, killing cores and RAM frequenzy.
Moreover, with the elderly 'OpenMediaVault'image from TK, my RK3288 never got lower than 600MHz. I don't know if OMV is so demanding.
I haven't tested this with the latest v4.11 nightly, I will have to do that.

 

scale governor; can you set this to interactive, I think currently it is at: conservative.

 

HDMI - Currently I am using mine headless.

 

I don't know if there sits a chip on the board of which we can read out the power consumption, up till now I had no issues with my Nexus 4 Micro-USB cable, but I haven't connected anything else but ethernet.

[Igor]

I notice HDMI signal black out from time to time without doing anything. (MiQi, kernel 4.11)

Wrote on mobile
 

[TonyMac32]

7 hours ago, Tido said:

1.8 GHz, I guess it is easy to replace the DTS in the build to keep running it at 1,6GHz. So it is fair enough to change this. On the other hand it would be interesting looking first at different throttling options like reducing frequency, killing cores and RAM frequenzy.
Moreover, with the elderly 'OpenMediaVault'image from TK, my RK3288 never got lower than 600MHz. I don't know if OMV is so demanding.
I haven't tested this with the latest v4.11 nightly, I will have to do that.

 

scale governor; can you set this to interactive, I think currently it is at: conservative.

 

 

My testing has shown it never going below 600 MHz, despite having op points for as low as 126 MHz.  I have only seen it go to 1.8 when trying to play video, since none of that hardware has userspace drivers yet to my knowledge (and I admit I don't know how to install them)  Rockchip has a repo for Mali and their MPP, but I'm not sure how to go about using them.

 

I think it is set to conservative/powersave, that question needs posed to a larger audience however, the kernel config affects both Tinker Board and MiQi.

[TonyMac32]

I am going to review powering the board through the GPIO, I am using a 2.4 Amp power supply to power it through micro-USB and am still getting a 400 mV voltage drop when the processor clocks up to 1.8 GHz.  (The supply is powering at 5.25V, I'm droping to just shy of 4.9)

[Tido]

IIRC, USB can go as low as 4,75V according to specification.

 

Maybe you already know this one..
linux/Documentation/devicetree/bindings/mfd/rk808.txt:
http://elixir.free-electrons.com/linux/latest/source/Documentation/devicetree/bindings/mfd/rk808.txt

 

I would be nice to understand how I can make mA visible just like I could with: regulator.4, vdd_arm
I tried to understand an looking for Ampere but seems to be only Voltage.

Datasheet for RK808, but only on the website in a Flash-Frame: http://download.csdn.net/detail/cc15673277477/9366343

[TonyMac32]

You're right, but I am using a pretty solid supply and cable, anyone using even an "ok" one is going to have some trouble.  (For instance, if it was a 5.0 or 5.1V supply, or the cable had any more inline resistance).

 

[Tido]

So it would be nice to read out the regulator RK808 (in idle & load), whether it gets no enough juice or you loose the mV at the connectors.

Where do you measure 5,25 /4,9V and how do you check this to the reference of the source under load?

I know it is your job, but I learned to always review /challenge results (I did something similar here)

 

[TonyMac32]

I'm reading the voltage at the GPIO 5V pins (Pins 4 and 6), which, according to what ASUS would call a "schematic", is directly connected to the same rail that sources the RK808.  This goes through a surge suppressor and a USB charge monitor, but it is the main 5V rail for the device.  The voltage was measured at that point with the CPU at idle and running minerd just to put some (albeit unrealistic) load on it (1.8 GHz).  The VCC_SYS can be observed to drop significantly.  I used a 5.0 Vsupply and the initial voltage, same test point, at idle, is 4.85 V.  For the 3.3 V components it's just fine, but USB peripherals will be less than amused.

[lucifercipher]

@TonyMac32

 

Hope this helps a little:

 

root@tinkerboard:~# lscpu
Architecture:          armv7l
Byte Order:            Little Endian
CPU(s):                4
On-line CPU(s) list:   0-3
Thread(s) per core:    1
Core(s) per socket:    4
Socket(s):             1
Model:                 1
Model name:            ARMv7 Processor rev 1 (v7l)
CPU max MHz:           1800.0000
CPU min MHz:           126.0000
BogoMIPS:              12.09
Flags:                 half thumb fastmult vfp edsp thumbee neon vfpv3 tls vfpv4 idiva idivt vfpd32 lpae evtstrm
root@tinkerboard:~# uname -a
Linux tinkerboard 4.4.16-00006-g4431f98-dirty #1 SMP Mon Apr 17 17:27:25 CST 2017 armv7l GNU/Linux

 

root@tinkerboard:~# cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_cur_freq
126000
 

Seems to throttle down to 126Mhz on idle.

[tkaiser]

53 minutes ago, lucifercipher said:

Seems to throttle down to 126Mhz on idle.

 

Armbian installs cpufrequtils and minimum cpufreq is set to 600 MHz with RK3288 boards. That's the only reason why it never goes below this value a few seconds after booting. Anyone is able to change that (/etc/defaults/cpufrequtils) but you should check your expectations or let's better say feelings first (why do you want to lower this value, what do you expect from it, how fast does the used cpufreq governor increases cpufreq based on load if necessary -- lower min cpufreq can lead to much lower overall performance if other settings don't fit -- and you should try to understand the meaning of a so called WFI instruction since there shouldn't be that much difference idling at 600 MHz or 126 MHz)

 

TL;DR: Most probably there's nothing wrong with 600 MHz

 

Edit: Forgot that we're talking about Cortex A17 here: http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.ddi0464f/CACJFAJC.html -- anyway unless one sets up a sufficient monitoring environment providing thermal and consumption numbers for both 126 MHz and 600 MHz and performance numbers for real world workloads (NOT full load benchmarks) staying at 600 MHz seems reasonable to me. If someone will spend some time on this checking through all available cpufreq governors with all available kernels is required to propose a change to Armbian's defaults  

[TonyMac32]

Thanks tkaiser, I hadn't been too worried about it, sounds easy to experiment for those who want to without changing anything in the defaults.

 

Along other lines, I tested a USB soundcard with the Tinker Board and can confirm that "as-built" the performance is terrible, as was mentioned earlier.  The default cpu governor is set to Powersave.

[traumfaenger]

Hello, folks. First of all, I really appreciate your efforts maintaining practical and powerful Armbian image for this board.

You, guys are the reason why I have returned my Banana Pi m2u back to the seller.

 

Today I got my Tinker in the mail.

 

I've installed Armbian 5.27 (stable) with kernel version 4.4.66, it is pretty smooth, and I'm not looking back for the BPi m2u.

 

Thank you very much!

 

I have attached a picture, of what I have found out, but you will surely know this already.

 

Tinker is running via SSH, hosting Pi Hole and some other network-based services. No GUI, no video acceleration required in my setup.

 

When connected to a Gigabit Network, I can write and read up to ~30MB/s to a Tinker-attached USB drive. (Compiled ntfs-3g latest stable 2017.3.22)

 

My question is, whether, and to what extent does the speed drop, when I do a copy to two drives connected to Tinker simultaneously.

 

I mean, USB/Ethernet bus is not shared, what is big pro. But its SoC has internally 3 USB hosts (info provided by dmesg, one of them is also OTG) and a USB Hub is connected to one of them providing us with 4 ports. The bandwidth of the USB drives would need to split between them because of the HUB. Can you agree on that?

 

In this scenario, I would copy a file to two network shares (Samba) simultaneously, the speed would drop to half. Resulting in 15MB/s for each drive. Have anyone tested this?

 

Also interesting is USB to USB transfer rate, since Ethernet is not involved in that process.

I mean, the USB 2.0 bus has 480Mbps of bandwidth, but ~30MB/s is about 250Mbps, so this is the real speed (overhead subtracted).

In addition, would be there any bandwidth free for the second drive in USB to USB scenario leading to the transfer rate to be the nearly same? Like if 130Mbps of 480 would be available to the second drive?

 

Again, thank you very much!

 

Edit: I have experienced following trouble: When copying at full speed (CPU usage is about 49%; Computer to Tinkers USB drive on Gbit LAN) it seems that the system has not enough internal bandwidth, or an ability to set an interrupt to be able to handle PiHole DNS.

Because of that behavior, pi hole does not get enough "CPU time" or whatever and does not resolve and forward a DNS address. I understand, it is busy by copying, of course, but it is be able to do multitasking, right? Since only 1/4 of total available Gbit speed is used by copying the data. It still should be able to do other network stuff.

[traumfaenger]

Well, I did some benchmarks regarding built-in USB Hub, Samba share write/read, and copying from USBHDD#1 to USBHDD#2.

 

First of all, the file system on the USB Drive made not much difference (about 2MB/s more or less).

Both drives are from WD, one is formated as NTFS (1TB, USB3.0), the other as ext4 (750GB, USB 2.0).

 

As long as I was doing a single write-operation through Samba to one of the USB drives, I got a transfer rate ~30MB/s.

Reading data from the drives by copying them to local machine was about the same speed ~30MB/s.

 

Now, when writing same file simultaneously to both drives, the bandwidth splitted by half for each drive, resulting in ~15MB/s for each drive.

Reading data by the same scenario was also ~15MB/s for each drive.

 

Interesting part was trying local (on Tinker, SSH, rsync) USB Drive #1 to USB Drive #2 copy and the other way around.

There, the choise of the file system gave +/- 2MB/s being somewhere between ~8-11MB/s.

 

As I did the benchmarking, the system was pretty stock only running Pi-Hole in the background, so pretty much all the resources were free.

CPU usage did not climb over 80% by the ntfs-3g driver, so the CPU (max. 1.6GHz on the image used by me) had enough potential.

 

The limiting factor is in fact the freaking USB Hub! As a member of this forum mentioned, ASUS made really stupid move by not soldering one more USB Port, which is provided by the SoC  However, the system is fluent, can deal with my upload speed, and makes fun! Maybe, someone would find this test interesting. Thanks

 

Edit: Just in case, someone wants to know... 5V/2A power supply did not managed to spin up one of the 2.5" drives properly, although the system was stable and did not crash. I've used the RPi3 official supply 5/2.5A, and it powered the board and two 2.5" drives well.

[TonyMac32]

Did you change the cpu governor before running the test?  I'd be curious your experience changing it to "interactive".

[tkaiser]

12 hours ago, Ruslan Dzanmahmudov said:

The bandwidth of the USB drives would need to split between them because of the HUB. Can you agree on that?

 

Correct, the Tinkerboard as well as the MiQi show here another serious design flaw (Micro USB for DC-IN is the other): Not exposing one of the two available SoC's USB host ports and connecting all USB receptacles to the remaining host port via an internal USB hub. So if you want to connect anything bandwidth intensive forget about both highest performance (hub) and to use more than 1 USB port (shared bandwidth). For such use cases Tinkerboard is simply an overpriced fail

 

For any further tests you should switch to 'performance' (/etc/defaults/cpufreq) now since otherwise you get numbers without meaning anyway with kernel 4.4 (since so far no one knows which cpufreq tweaks Rockchip engineers implemented here)

[lucifercipher]

37 minutes ago, tkaiser said:

For such use cases Tinkerboard is simply an overpriced fail

 

And i bought two of this overpriced fails. That feeling when you experience after a screw up, i'm there. 

[traumfaenger]

1 hour ago, lucifercipher said:

And i bought two of this overpriced fails. That feeling when you experience after a screw up, i'm there. 

Well, I think this board is better than a Banana Pi m2u, which has the power, but the software (and somehow the power management) is such a non-sence! Apart from this, as I took the board in my hand (BPi), I thought I would crack the PCB in half.

Thank god, people over here are contributing to this board / chipset.  It's simply hard to find (well, at least it was for me) a decent board with decent specs, which is also available at the retailer.

 

@TonyMac32 No, the cpu governor was the default shipping with Armbian 5.27 (kernel v. 4.4.66, CLI) -> Tests have been done with "conservative" governor. During the day I'll make the test again with 'interactive' & 'performance'

 

@tkaiser Thank you also for your feedback. The system is still not completely set yet, so I can play around. And what about trying out the 'NEXT-branch', there are located the mainline kernels. I think will try this after USB bus test.

[tkaiser]

2 hours ago, lucifercipher said:

And i bought two of this overpriced fails. That feeling when you experience after a screw up, i'm there.

Well, I fail to understand this 'decision' every single really fast SBC is plagued with. Where's the problem putting 3 USB receptacles behind the internal USB hub and connect the 4th to another receptacle that's directly connected to the SoC's other host port (and call it 'disk port' or use another color to identify the port that has not to share bandwidth)? But we see both failures also on MiQi board, on NanoPi M3 and on the worst SBC available: Banana Pi M3. At least on those 2 M3 DC-IN can be fixed/workarounded somewhat.

 

On ODROID-XU4 it's the same but at least we're talking about USB3 there (though the hub in between has also some potential for causing troubles, eg. cutting SuperSpeed data lines after a bus reset and stuff like that)

 

1 hour ago, Ruslan Dzanmahmudov said:

I think this board is better than a Banana Pi m2u

 

Of course but that's no surprise since every Banana released after BPi, BPi Pro, BPi M1 and M1+ was simply utter crap plagued by stupid design decisions, instable software and non-existent vendor support. We could add the original M2 (A31s based) to the exception list since this board runs stable at least with mainline kernel but USB situation is also driven by plain stupidity (4 receptacles all behind an internal USB hub leaving the SoC's other USB host port unused -- what do 'engineers' think when doing this: 'F*ck you, silly customers?').

 

When you do further testing it would be interesting to run through every available cpufreq governor while executing this here (`pwd` being on SD card):

iozone -e -I -a -s 100M -r 4k -r 1024k -i 0 -i 1 -i 2

 

[Tido]

57 minutes ago, tkaiser said:

(and call it 'disk port' or use another color to identify the port that has not to share bandwidth)

 

RK3288 A17 - comes with 3 USB ports.
2 x Standard
1 x OTG

If the drawing from Ruslan ist correct two standard are used for Hub & Audio,  then only the OTG is left. Can you run UASP over OTG ?

 

And if it is possible to do so, I guess they either didn't think about it OR it was too expensive.

[tkaiser]

1 hour ago, Tido said:

Can you run UASP over OTG ?

 

If you're asking me: no idea, no interest. If you're interested in you need to find out which USB IP is used on this SoC, which driver deals with it and so on.

 

As already said: for most of my use cases this Tinkerboard is just an expensive fail. If I've 60 bucks I buy 4 x NanoPi NEO2 and not 1 Tinkerboard. All those H5 boards like Orange Pi PC2 or Prime, NanoPi NEO2 or NEO Plus 2 soon have 3 real USB2 host ports exposed that do not have to share bandwidth and the Micro USB port there can be switched from OTG (own controller) to a real USB host port (using an own OHCI/EHCI controller pair) so you've 4 real USB2 host ports there.

 

And even  $15 NEO2 has Gigabit Ethernet on an own bus. And unlike giants like ASUS/Rockchip vendors listen to community: http://www.cnx-software.com/2017/05/12/nas-dock-kit-v1-2-gets-support-for-nanopi-neo-2-an-uas-capable-usb-to-sata-bridge-and-an-rtc-battery/ (just 5 weeks between first version that sucked somewhat and created some public complaining by community and the fixed version now)

[traumfaenger]

9 hours ago, tkaiser said:

When you do further testing it would be interesting to run through every available cpufreq governor while executing this here (`pwd` being on SD card):

iozone -e -I -a -s 100M -r 4k -r 1024k -i 0 -i 1 -i 2

@tkaiser What is it for? I have a really slow SD card from sandisk, class 4. Does it measure the File IO performance om the sd-card or through-out the whole system?

Of course I will  replaced my old SanDisk by a UHS 1 card, but as long as I am playing around I will leave it "as it is".

 

Now, I'd like to share more benchmarking regaring governor... I did my best, tryed to be as precise as I could be

[traumfaenger]

Hello, folks, it's me again.

I have successfully installed the NEXT-build of Armbian 5.27 with Kernel version 4.11.0-rockchip.

So far it is running smooth, but I get an annoying bug. I've read that you guys are aware of this one.

 

When I do restart my system using "sudo reboot" the system does not start again. It's like it enters the "halt state" and remains there till power loss.

I've read that there is a patch for 4.4.66 in this thread. Will there also be on for 4.11 available? Thanks

 

[TonyMac32]

The mechanism used to get the reboot working on 4.4 does not work on 4.11.  I haven't dug into why exactly that is (it is a hack), however I saw some soft reset patches to the 4.12 kernel that may resolve the issue, rock chip has been pretty active with improving kernel support, although the RK-wifi system interface isn't present for wifi, etc.

[traumfaenger]

Thank you very much

Well, it's a small and powerful SBC which shows compatability to RPi, and I think it will attract people to deal with this platform.

These are also kinda good news, because work is being done. What I hope, apart of you guys here @Armbian,

is the fact, that ASUS is not a small non-sence China corp. So I hope they will deliver

 

Has anyone experienced "dnsmasq", which is present, but is not working properly. When I restart the service, it will come back online,

but again, after coule of mins my browser can not resolve IP address. Pi-Hole uses it to block ads on DNS-level.

Update: Pi Hole seems to be running fine. I'm excited to test this box with Seafile  Maybe I'll get over 2,3MB/s as it was the case with RPi3

[tkaiser]

13 hours ago, Ruslan Dzanmahmudov said:

What is it for? I have a really slow SD card from sandisk, class 4. Does it measure the File IO performance om the sd-card or through-out the whole system?

 

The iozone test should just give an indication how the different governors weigh IO activity (only useful with Rockchip's 4.4 kernel since with mainline it's known). But no need to continue, I'm not interested in Tinkerboard at all any more (the first RK3328 based board is just around the corner featuring USB 3.0, dual Ethernet including one Gigabit Ethernet MAC and not using the most shitty DC-IN connector unlike Tinkerboard).

 

BTW: If you think ASUS will deliver... they won't since they're not even involved in development (all done by Rockchip engineers and in the past by the communities around Firefly-Rk3288, Radxa Rock2 and MiQi before) and it takes them weeks to (still not) fulfill their promises (sending out dev samples).

[traumfaenger]

Dude, I like you

 

Seriously, I agree with you, but there is something what I won't do, simply because of the use-case I am up to. You see, you are right, 60€ and not even a proper SoC to USB output, I mean, there have to be soldered 2/4 lines from the SoC to USB to be able to provide fully operational port, but they didn't.

 

But what expectations do I have. What will I be using it for. What is going the board you are talking about cost?

 

For me, it's about a fast CPU, to handle the network requests. I want fast Adblocker (PiHole), own DNS Service with local caching, openvpn to be able to connect to my data on network, and simple cloud solution That's all. And for this scenario I think this board has still more potential than RPi3

 

I had this awful Banana pi m2u whick looked hardware-wise indeed very promising, but it's software was a no-go.

And if I wait for a board, I mean, we all know PC industry. You can basically wait all the time, and you get better and better devices for your money.

 

And that's why I'd like to thank everyone who is contributing to this project I have basic knowledge of C# and Cpp but it's by far not enough to slip into kernel-thingy. 

 

I am actually pretty happy when I can compile some user-space driver (ntfs-3g)

[Da Alchemist]

2 hours ago, tkaiser said:

(the first RK3328 based board is just around the corner featuring USB 3.0, dual Ethernet including one Gigabit Ethernet MAC and not using the most shitty DC-IN connector unlike Tinkerboard).

 

 

 Who is bringing us this Board? When it is orange pi , i hope there will be the 8 channel I2S interface... (just a Dream ).

Btw, is there 8channel Audio out via HDMI on the TinkerBoard?

 

Regards

[jernej]

4 minutes ago, Da Alchemist said:

Btw, is there 8channel Audio out via HDMI on the TinkerBoard?

 

4.4 kernel should support it since last month or something like that (I was analyzing exactly that piece of the code), but I didn't test it.

[tkaiser]

2 hours ago, Ruslan Dzanmahmudov said:

Dude, I like you

 

BTW: I don't want to discourage anyone working on this board and further improving support. It's just me giving up on this hardware since ASUS performs so lousy (wrt deal with 3rd parties wanting to do exactly that: improving software support for their hardware). So as a last hint: since I believe the only really interesting use cases for this board are 'Desktop replacements' you should focus on this when experimenting with different cpufreq governors. Most desktop tasks benefit a lot from cpufreq being ramped up as quick as possible. Also have a look at: https://forum.armbian.com/index.php?/topic/4246-can-improve-the-desktop-performance/ (though I believe it's also important to improve IO behaviour of browsers)

 

 

14 minutes ago, Da Alchemist said:

Who is bringing us this Board?

 

This is Rock64 with an I2S companion board sitting on it: https://drive.google.com/file/d/0B0KJDZUkcqOqUnBYNU9JN0EwV2c/view (details will be available soon so I won't share more information now since there isn't much more I know anyway  )