Announcement : Odroid N2

[NicoD]

23 hours ago, NicoD said:

Many people get their info on Youtube, it is mostly another public then we meet here in the Armbian forum

 

1 hour ago, chwe said:

I seriously don't read many benchmarks.. I bought my notebook without looking at any benchmark.. I decided

 

You, @chwe the people
I do watch a lot of benchmarking on Youtube pure out of entertainment. I love Gamers Nexus on Youtube. I don't care a bit about x86 hardware. I'm not planning to buy anything new. But it seems to have an entertainment value that I like.
 

2 hours ago, chwe said:

But you target audience is more the end-user side right? So it might be interesting for them what they can expect by buying *random board* from *random boardmaker*.

Indeed, a lot of people who are new to SBC's. So I try to keep it as simple as possible, and show how to set up everything for a new user, and what's the best OS available, gaming, video playback, energy consumption, ease of use, ... My viewers seem happy, I almost never get bad comments. I many times get the comment that I've saved them days of searching. So that's what I keep doing.
I try to test as much as possible, and only when I don't know anymore what more to test I make my videos.
Most other sbc reviewers on Youtube make their video after only a few hours of working with them. That's where I try to be different. That also means my videos are later than the more popular Youtubers.(ETAPrime can't even install a simple Linux distro on an sbc, and he's got +200 000 subscribers)
I now get boards for free, and I share my findings with the board makers. FriendlyElec and Radxa have done a lot with my data for the M4 and RockPi4B. I'm happy with that. Other makers don't seem to care as much.
I also don't make a video as long as a novice user can't use a board easily and well.
 

 

2 hours ago, chwe said:

How often did you run those tests? Roughly 25% looks indeed like a real difference but I would prefer to have at least a triplicate to get a clue how much error margin I expect for such a test. I'm quite sure you'll still see a difference but the dataset locks not complete (e.g. 'RockPro64, Ubuntu Cosmic with LXDE, 2.0/1.5GHz, CONFIG_HZ=1000: 1:01:11' is missing)

I've done a lot more of them with the RK3399. When settings haven't changed Blender has very little changes when done multiple times. (I'll do 3 now with the M4)
Ignore gimp, gtkperf and sysbench. That was just to show you're nothing with that.
I don't think the Rock Pi 4B had CONFIG_HZ=250 but HZ=1000. The NanoPi M4 now did 5minutes better than the RPi4B with HZ=250 at the same clockspeed. So my assumption that the RockPi4B was faster because of faster ram is untrue.  Still need to do more with the RockPi4B. I'm waiting for the Armbian images to mature.

Rock Pi 4B             |SBC bench result  |CPU Miner          |7-zip s/c |7-zip b/c     |7-zip multi avg. of 3 |Blender     |GIMP    |GTKPerf |Sysbench
Ubuntu                  http://ix.io/1uVr  9.50kH/s            1242       1818           7802                   1h17m22s

NanoPi M4              |SBC bench result  |CPU Miner          |7-zip s/c |7-zip b/c     |7-zip multi avg. of 3 |Blender     |GIMP    |GTKPerf |Sysbench
Armbian bionic hz1000   http://ix.io/1nLh  10.23kH/s           1335       2005           8352                   1h13m50s     0m29s5   5.06s    26763
 CONFIG_HZ=250          http://ix.io/1BLW  10.45kH/s           1335       2007           8320                   1h08m28s
Armbionic@1.4/1.8 hz250                                        1253       1828           7821                   1h12m52s

Armbian bionic nightly  http://ix.io/1pDo  10.24kH/s           1329       1990           8292                   1h13m28s     0m29s    5.12s    26733
Armbian stretch desktop http://ix.io/1odF  8.66kH/s            1350       1977           8400                   1h14m12s     0m31s    5.24s    3.1s 
Armbian stretch dsk nightly  //ix.io/1pM0  8.80kH/s            1359       1993           8500                   1h15m04s     0m31s    5.32s    3.3s 
Armbian stretch core no fan  //ix.io/1pKU  8.80-8.65kH/s       1353       1989           8461
Armbian stretch core         //ix.io/1pL9  8.76kH/s            1354       1988           8456
Armbian stretch core nightly //ix.io/1pLf  8.82kH/s            1357       1994           8494
Lubuntu Bionic arm64    http://ix.io/1oGJ  9.24kH/s CPU Miner  1056       1551           6943                   1h28m13s             
Lubuntu Bionic armhf    http://ix.io/1pJ1                      1111       1769           7705                   2h02m54s     0m57s    6.97s    1666      32-bit
Lubuntu Xenial armhf	http://ix.io/1oCb                      989        1507           6339                   2h20m51s     0m59s    49.77s   49.7s     32-bit

 

2 hours ago, chwe said:

his approach differs from @tkaiser benchmarking.

 

Of course @tkaiser has a lot more knowledge than me, and I'd rather see someone like him making good videos to well inform people. As long as that doesn't happen, there's still a place for me on Youtube (TKaiser, the Derbauer of SBC's??)
I don't think he'll ever aprove using Blender, but I also use SBC-Bench. I don't think there's much to say against that. 
Blender does have a few advantages for me. It shows if a system is stable(crashed if not), it takes a long time so I can see the heat characteristics, it uses close to 100% of all cores(7zip doesn't, and this differs a lot from board to board, distro to distro). But small changes can effect the performance a lot.

@rooted I'm also interested in seeing how the H2 does against the N2. Also power consumption while maxed out interests me in both. Until now the NanoPC T3+ has the lead in horsepower per watt for me(I need to retest that with HZ=250 too). I think the N2 can beat it.
But time enough for that. Take good care of each other. May everything be well and safe for your son.
 

Quote

We create our own future, if you firmly believe that you will be all right-then everything will be fine.

That's a nice filosofy to live by, but we don't have faith in our own hands. Health is something we can only be thankful for having it.

 

[rooted]

40 minutes ago, NicoD said:

I'm also interested in seeing how the H2 does against the N2. Also power consumption while maxed out interests me in both. Until now the NanoPC T3+ has the lead in horsepower per watt for me(I need to retest that with HZ=250 too). I think the N2 can beat it.
But time enough for that. Take good care of each other. May everything be well and safe for your son. 

 

I have the means to see power usage, as time and situation permits I will test this as well.

 

Thanks for the kind words.

[NicoD]

7 hours ago, chwe said:

but I would prefer to have at least a triplicate to get a clue how much error margin I expect for such a test.

The 2nd time I didn't reboot before starting the test again.
Then the 3th and 4th times I everytime started with a clean boot(as the first of course). I'll do it again without a clean boot now, just to see if I can recreate the 2nd result.
But it's all within a minute of each other. And 1st 3th and 4th even within less than 10 seconds in a +1 hour test. So I think that's quite reliable.
I never look at differences of less than a minute, since that's a too small margin.

 

[chwe]

5 hours ago, NicoD said:

9 hours ago, chwe said:

I seriously don't read many benchmarks.. I bought my notebook without looking at any benchmark.. I decided

 

You, @chwe the people

if I would speak for others I wouldn't write I.. right?

 

5 hours ago, NicoD said:

I don't care a bit about x86 hardware. I'm not planning to buy anything new.

well, most of my serious work depends on windows, ergo depends on a flawlessly working x86 machine.. software support for the tools I need is nowhere either on arm and/or linux (something simple as a chemical structure editor which fulfills my needs simple doesn't exist on linux and I'm not a masochist using the few structure editors available which are open-source).

 

1 hour ago, NicoD said:

But it's all within a minute of each other. And 1st 3th and 4th even within less than 10 seconds in a +1 hour test. So I think that's quite reliable.
I never look at differences of less than a minute, since that's a too small margin.

Or you calculate some statistics without meaning to let it look like a professional..

 

If I'm not completely wrong it should be:

x_mean = 01:07:27.56 (+-6s)

std. deviation = 24s

 

Assuming it's a Gaussian type distribution. Or otherwise called: the easy way to look results more trustfully even if there's absolutely no reason to do so!

[NicoD]

I've got my N2. It's a beast with a beast of a heatsink.

It outperforms everything else that I've got by miles. And it will never overheat. No fan needed with this.
The maximum stable OC is 1.992Ghz/1.992Ghz. There's an option for 2.004Ghz for the A73 that isn't stable anymore.
At OC maxed out for long it never went over 61°C and probably will never go any higher. So the reasoning that it will heat up the board too much doesn't fly.

I'm going to do a lot of Blender renders the first weeks with it. So I've done my Blender test to see how it compares with other SBC's.

Default 48m40s
OC         46m02
 

The best for my NanoPi M4 is 1h08m28s   (1.5Ghz/2Ghz    CONFIG_HZ=250 ) 
NanoPC T3+                             1h10m25s

It's going to do a great job for me.
For desktop use it isn't perfect yet. Still a few small issue's with Ubuntu. No VPU acc yet. But for an SBC that's as new as this it's already amazingly good.

It's lighter, and smaller than I had imagined. The pictures I've seen were all bigger than the board in realit.

I would have liked to have 5Ghz wifi on it. It's not of this time to not have wifi.
Done with having my PC blocked the whole time while rendering.
Now I just need something to build Armbian. Maybe the Odroid H2 next? Or would it be the NVIDIA Jetson Nano?

[TonyMac32]

Remember the Amlogic blobtastrophy. Your board may never have hit 1.9 GHz.

Sent from my Pixel using Tapatalk

[NicoD]

4 minutes ago, TonyMac32 said:

Remember the Amlogic blobtastrophy. Your board may never have hit 1.9 GHz.

I'll try to inverstigate on that. It does perform great.
The governor says 1.99Ghz / 1.91Ghz while it should be both the same. That's something strange.
Now doing my first 4K super high quality blender render. After that I'm going to inverstigate a bit more.
I haven't red that much yet about it, I didn't know there was an issue with this.

ps. Blender crashed, so I guess it's not stable enough at this OC. A bit less it'll be then...

[TonyMac32]

Hmm, if it's unstable, perhaps Odroid has again gotten a "real" dvfs blob. Search "Amlogic clock cheating" in the forum.

Sent from my Pixel using Tapatalk

[NicoD]

4 minutes ago, TonyMac32 said:

Hmm, if it's unstable, perhaps Odroid has again gotten a "real" dvfs blob. Search "Amlogic clock cheating" in the forum.

I did overclock the impatient way. Immediatly the highest possible clocks. Now going down one by one. The highest wasn't stable for anything.
I had just changed the clocks while the governor didn't show a difference. It crashed again. This was with the A53 at max of 1.992Ghz and A73 at 1.908Ghz. Nothing in between that and the default clock for the A53. So default it'll be for the A53. Maybe the A73 can clock higher while stable. Not that I need it overclocked. It's fast enough stock, and it doesn't make a too big difference. But I must know what's stable and what's not.

I do know the story of the C2 and the 2Ghz claim.  I would not expect the people of Hardkernel fall for that trick twice. It does perform great, can't cheat with that. I do need to find out how an A73 core compares to an A72 core at the same clockspeed.

[AndrewDB]

4 hours ago, NicoD said:

Done with having my PC blocked the whole time while rendering.

@NicoD I noticed you have benchmarked it against other SBC's using blender, but how long does your PC take to render the same file and what kind of CPU does it have?

[NicoD]

27 minutes ago, AndrewDB said:

I noticed you have benchmarked it against other SBC's using blender, but how long does your PC take to render the same file and what kind of CPU does it have? 

My pc does 22 minutes over it. It's a I5 2500K clocked at 4.3Ghz. It's an old cpu, but the arm boards still don't reach it. But the N2 fast enough for what I need it to do.

[Poincare]

Hello, just curious ... is there a working Armbian for folks with the N2 ?  I didn't see a release on the download page and forums search didn't find it, if it exists.  Thanks.

[jeanrhum]

On 4/28/2019 at 8:00 PM, balbes150 said:

New image DEV for Odroid N2  with kernel 4.9 (regular kernel from HK). The sound works and everything else should work (I didn't check).

Please note that in this version the functions are managed in the file "boot.ini"

 

It's not an official armbian image, but balbes150 custom images based o armbian build tools than can be downloaded from here:

 

[TonyMac32]

Hello, just curious ... is there a working Armbian for folks with the N2 ?  I didn't see a release on the download page and forums search didn't find it, if it exists.  Thanks.

Besides the great work of Balbes, I am the primary caretaker of Amlogic on Armbian, and I'm afraid I don't have an N2.

Sent from my Pixel using Tapatalk

[umiddelb]

5 hours ago, TonyMac32 said:

Besides the great work of Balbes, I am the primary caretaker of Amlogic on Armbian, and I'm afraid I don't have an N2.

I can get in touch with the odroid guys to change this situation if you want.

[martinayotte]

4 minutes ago, umiddelb said:

I can get in touch with the odroid guys to change this situation if you want.

Perhaps me too ...

[TonyMac32]

I can get in touch with the odroid guys to change this situation if you want.

Thank you for the offer, I wouldn't want anyone calling in favors they don't need to. There is a very high likelihood this won't take much effort/maintenance thanks to Bay Libre and obviously Balbes, I haven't been pushing for a board because I've got quite a few of them that still need my attention, I'm hesitant to ask for a board I might not get a lot of time to work with (looks at RockPi and Renegade Elite)

Sent from my Pixel using Tapatalk

[sfx2000]

On 4/10/2019 at 7:57 AM, NicoD said:

I've got my N2. It's a beast with a beast of a heatsink.

 

Would be interesting to see how it competes with Jetson Nano with a good power supply...

 

Jetson has a fairly large heatsink, but it's generally cool to the touch right now with 2.5A over MicroUSB - 5 amp Power Supply is scheduled to arrive tomorrow, and then we can turn up the Jetson clocks and see what it can really do...

 

(Jetson Nano under full load can easily consume 20w when doing GPU compute tasks - not too much different than Google Coral)

[balbes150]

On 5/2/2019 at 10:47 PM, martinayotte said:

Perhaps me too ..

 

On 5/2/2019 at 10:49 PM, TonyMac32 said:

Thank you for the offer, I wouldn't want anyone calling in favors they don't need to. There is a very high likelihood this won't take much effort/maintenance thanks to Bay Libre and obviously Balbes, I haven't been pushing for a board because I've got quite a few of them that still need my attention, I'm hesitant to ask for a board I might not get a lot of time to work with (looks at RockPi and Renegade Elite)

 

Soon to appear interesting device S9xxx . If you are interested, I will try to negotiate with the management of the company to send you samples.

 

PS According to preliminary information, it should be a very interesting device.

[NicoD]

12 hours ago, sfx2000 said:

Would be interesting to see how it competes with Jetson Nano with a good power supply...

  

Jetson has a fairly large heatsink, but it's generally cool to the touch right now with 2.5A over MicroUSB - 5 amp Power Supply is scheduled to arrive tomorrow, and then we can turn up the Jetson clocks and see what it can really do...

It all depends on what you want to compare. Comparing the CPU is the easiest. In my opinion it's best to use 7zip single core benchmark(since 100% use) and multiply that by the number of cores at that speed. I only use decompression numbers because I do not want to mix compression with decompression.

The Nvidia Jetson Nano has 1421 single-core x 4                                                         = 5684 (approximation of it's maximum possible CPU computation power)

The Odroid N2(no OC) small cores : 1647 x 2 = 3297 + (big cores 1905 x 4 =   7620) = 10917 (again approximation, just a number...)

The NanoPi M4 small cores : 1335 x 4 = 5340 + (big cores 2005 x2 = 4010)               =  9350 (....)
The NanoPC T3+ : 1290 x 8 cores                                                                                  = 10320  (....)

To compare GPU power is a lot harder. And most important about SBC's is drivers for it, and in that case the NVIDIA Jetson Nano probably outclasses the N2. The N2 doesn't have good GPU/VPU drivers for Linux.
The N2's GPU is also very powerful, but can't be used the way the Nano's GPU can be used. I've got a new USB3 4K camera. The N2 can't handle more than 720p 60fps software encoding. While the Nano would have no problems with 4K hardware encoding. And even multiple camera's at a time.
You can run a Android benchmark to get some sort of GPU score. But it doesn't say anything about the useability in Linux.

For that the numbers aren't that important, but how well you can use it for the goal you've got for it. I now could use the Nano with the USB3 camera. Would have made life a lot easier to review the camera.
I would have loved to buy the Nano. But it was that or the N2. Now saving money for a big trip. Maybe after that I'll see to source one.
The N2 does excell in power consumption. Only about 6W maxed out(only CPU). That's lower than some quad-core SBC's. Probably the 12nm that's the biggest cause for this.
 

4 hours ago, balbes150 said:

it should be a very interesting device.

Thanks to your software it's an amazing SBC for me. I had 3 weeks of troubles and bad luck with the Odroid OS'es. I installed your Bionic 4.9 DEV and haven't had any issue. Haven't shut it down since I installed Armbian.
Do you know anything of comming availabillity of X11/VPU drivers? That would be a gamechanger for it. I hope it's going to be in mainline too.
The N2 could be an awesome Linux desktop "PC" for many. I don't know many Linux desktop tasks that need more CPU power than this. Now the NanoPi M4 is still better for that(and for NAS).

And here a link to my review video about the Odroid N2 for those interested. NicoD - Odroid N2 Full Linux Review

[sfx2000]

2 hours ago, NicoD said:

The Nvidia Jetson Nano has 1421 single-core x 4                                                         = 5684 (approximation of it's maximum possible CPU computation power)

The Odroid N2(no OC) small cores : 1647 x 2 = 3297 + (big cores 1905 x 4 =   7620) = 10917 (again approximation, just a number...)

The NanoPi M4 small cores : 1335 x 4 = 5340 + (big cores 2005 x2 = 4010)               =  9350 (....)
The NanoPC T3+ : 1290 x 8 cores                                                                                  = 10320  (....)

 

Normally with big.LITTLE, tasks migrate from the LITTLE core to the big core - much depends on the kernel scheduler in use, and whether that scheduler is aware of ARM MP in a big.LITTLE context - in any event, one does need to consider performance per core, and how to schedule tasks. Done right, it works pretty well, done wrong and one can get into contention conflicts between cache, main memory, and arbitration on who has control of the CCI at any given time.

 

nVidia also has the 4 Cortex-A53's along with the 4 Cortex-A57's, but they don't expose the A53's directly to the kernel - management is done below the kernel in the hardware, which might make scheduling easier in the linux kernel.

 

Worth reading - it's a few years old - https://www.arm.com/files/pdf/big_LITTLE_Technology_the_Futue_of_Mobile.pdf

 

More recently, there's been a lot of work around energy aware scheduling, and ARM is one of the big players there - https://developer.arm.com/tools-and-software/open-source-software/linux-kernel/energy-aware-scheduling

 

It's in Kernel 5.0, but perhaps some folks might consider back-porting it if they have the spare time to do so.

 

Anyways - more cores doesn't mean faster - there's a limit there - and this is more of a systems engineering issue, along with SW... and this goes back to 1967 with Amdahl's presentation...

 

https://en.wikipedia.org/wiki/Amdahl's_law

 

Much like Shannon with Information Theory and Wireless Communication - one can only put so many bits into a medium before it becomes error limited.

 

Anyways - it's an interesting time where we're getting the big ARM cores in boards that are $100USD or lower - but it also starts to point out that we need to have better solutions to power these larger cores...

[sfx2000]

Anyways - would be fun to see a cage match between Nano Pi M4 vs. Odroid N2 vs nVidia Jetson Nano...

 

Traditional benchmarks as well as some VPU/GPU work - as RK3399 has Mali T864 vs. Amlogic with G52 vs. nVidia with Maxwell

 

Since all are supported in some way with Ubuntu 18.04LTS with Vendor supplied BSP's...

[NicoD]

8 minutes ago, sfx2000 said:

Anyways - it's an interesting time where we're getting the big ARM cores in boards that are $100USD or lower - but it also starts to point out that we need to have better solutions to power these larger cores... 

Indeed. I love it to have these awesome boards for this price. The possibillity's are limitless with them. I didn't think 2 years ago we would have gotten this far. The XU4 was also a powerhouse. But these arm64 boards are just something else.
The N2 does have a good power solution. It's powered with 12V barrel jack. I've never seen such stable voltages on the 5V lane on an SBC. But it's got it's issue's as well.
And yes, multi-core performance is only important for applications that can make good use of it. Many applications still favor single core performance.

[sfx2000]

8 minutes ago, NicoD said:

Indeed. I love it to have these awesome boards for this price. The possibillity's are limitless with them. I didn't think 2 years ago we would have gotten this far. The XU4 was also a powerhouse. But these arm64 boards are just something else.
The N2 does have a good power solution. It's powered with 12V barrel jack. I've never seen such stable voltages on the 5V lane on an SBC. But it's got it's issue's as well.
And yes, multi-core performance is only important for applications that can make good use of it. Many applications still favor single core performance.

 

Much depends on execution - one of the first common boards with Big ARM is Tinkerboard - which was tragically limited by power and thermals and didn't reach the potential - out of box, RPi3BPlus actually outperforms it - it's not a bad board, but it has its limitations out of box.

 

Sounds like N2 has addressed some of these issues - better cooling, better power...

 

Jetson Nano with a proper power source (e.g. barrel not uUSB) - Jetson powered by uUSB is power limited to the point of stability unless running in low power model, which cuts off half of the CPU's and downclocks the GPU to stay within the 5W power budget, and this is one of the popular issues over on nVidia's support forums - Jetson suddenly powers off - and that's because that big GPU can ramp very fast - hence the 20W real budget there.

 

 

[balbes150]

12 hours ago, NicoD said:

Do you know anything of comming availabillity of X11/VPU drivers?

Drivers are in development but it won't be fast.

 

 

[sensei73]

Hi, there is no /dev/amstream* in Armbian so Kodi can't use hardware decoder. Could support be added please?