27 thoughts on “Odroid N2 Full Linux Review – The fastest SBC I’ve ever seen

  1. Cool video! I was wondering how the Odroid N2 does for laptop/desktop type usage. For example, how well does it handle using firefox with 10+ tabs along with youtube, etc and whether that would still leave some room for things like running a few lightweight server processes in the background. I suppose RAM at 4GB might become a bottleneck at more than 5-6 tabs in firefox?

  2. I have purchased also the odroid N2.
    Which version of Armbian do you use?

    I use now Armbian_5.90_Odroid-n2_Ubuntu_bionic_default_4.9.182_20190702.img (no need for the desktop version), but i have some issues.
    If i use the command: dmesg, i will get a lot of: drivers/amlogic/efuse/efuse_hw64.c:_efuse_read:196: read error!!!
    On other versions from forum.armbian.com, i have issues that temperature cpu does not show, with the command htop

  3. Great review! Lots of benchmark comparisons against other SBCs. Few people have those resources available. Very helpful as a buying guide.
    Odriod N2 looks great… but no NVMe SSD connectivity 😒. I'd snatch one up, if it had M.2. As you mentioned SSD via USB3 is limited. eMMC is an improvement over uSD, but neither are as fast as NVMe. Guess I'm spoiled with the Samsung 1TB SSD. If I'm buying an up market SBC, it's a requirement.
    I like the 12 VDC power input. Higher voltage = lower amperage. Power consumption is equal. Very compatible with automotive and many solar/storage options. As you know, I'm not a fan of USB-C power input, unless there's an on-board header to bypass the flimsy SMD connector. This has also become one of my requirements. Besides USB-C can be much more than a power input.
    So, that's TheHermit's humble opinion (most of us have one). 😎
    By~ye!

  4. Your reviews are the one of the best, I like how you focus on day to day tasks instead of just emulation and gaming.

    Im a moderate Linux user but I have not used an SBC in the past. Which board should I buy first? Will the raspberry Pi 4 be good? My PC has a core 2 duo 2.6ghz. I want to learn python and coding. Nothing too intensive. I'm also looking at this board and the nano Pi M4/ nano pc t4.

  5. wow, what I did not find on other channels: Blender benchmarks on those sbc's. just a suggestion: try to compile from source code Godot Engine to see how it performs. keep the good work.

  6. A pity about the 4k playback at the moment, but I'm glad it doesn't have wifi and BT. Not everything needs it, and not everyone wants it. By leaving it to an add-on you can add whatever version you want for a very small cost.
    More important is the 4k video, gigabit network and USB3 ports.

  7. Thanks for the review, I like how down to earth it is, very informative. Too bad about poor 4k playback in Firefox and USB 3.0 interference issues, but when I consider buying a product that's exactly the kind of information I am looking for. Other channels tend to omit such issues.

  8. Thanks again for another great review. 1st YouTube video I've seen @ 1440p HD. Nice opening graphics, too. Having difficulties choosing between UDOO x86 (waiting for Bolt release), LattePanda, NanoPC T4, or NanoPi Neo4 (so far). So, I'm spoiled for choice🤷‍♂️
    Yeah, I know. Different boards for different applications, with big price differences. Wallet says,"Go cheaper." Mind says, "NVMe." 🤑

  9. Nico, I did some follow up testing to make sure that the Jetson Nano could handle the load of simultaneous GPU-CPU rendering by opening and running TWO copies of Blender 2.79 and running TWO COPIES OF THE BMW27 BMW27 BENCHMARK (one with CPU and one with GPU).

    With BOTH renders running at the same time the Jetson Nano completes the times were:

    Jetson Nano CPU render: 1:08:35
    Jetson Nano GPU render: 1:03:47

    Note that this was with both renders set to a non-optimal 160×160 tile size to simulate the constraints of dual rendering in Blender, but even with this limitation, the combined performance is MUCH better than the Odroid N2. When you convert the above times into frames /hour and combine them, it works out to an equivalent per frame render time for BMW27 of just over 33 minutes !

    Of course, we will have to wait for Blender 1.8 to get the full benefits of both eevee and combined CPU+GPU on the Jetson Nano, but as I understand it, some of the latest 'daily builds' of Blender 2.79 currently already support combined CPU+GPU rendering – and even without this, my results above show that for time consuming animation rendering (like the opening to this video) you can ALREADY benefit by simply splitting the animation frames up into two batch jobs.running TWO copies of Blender (one set to GPU and one set to CPU). Taking your best 'overclocked' benchmark of 46 minutes and dividing by the combined score above of 33 minutes means that with the frame renders split between CPU and GPU the opening animation you showed at the being of this video would have rendered almost 40% faster!

    Also, when Blender 1.8 becomes available on the Nano, you can render overtly CGI scenes like your opening animation that make no attempt to be 'photo realistic' with Eevee at like 1000 times the speed of a full Cycles render – but I wouldn't expect Eevee to work on your N2 since it relies very heavily on proper up to date OpenGL support. Nvidia has best-in-class OpenGL support libraries argued by many to be the best available in Linux, where the Odroid N2 relies on the crappy non-hardware-accelerated out-of-date Mesa CPU based drivers.

    I'm sorry Nico, the Odroid N2 may have some use-cases where it beats out the Jetson Nano, but certainly NOT for Gaming, Blender, Video, or full featured desktop use – and for a file server, I wouldn't go with either a Jetson Nano or N2 but would look for a board like the Odroid H2 with SATA, and/or M.2 support since running storage on USB3 has been known to be flaky on some ARM boards due to flaky drivers.

  10. Thanks for another great review. I was looking at the N2 but decided to go with the Jetson Nano instead because of the better GPU driver support on the Linux Desktop. Desktop performance on the Jetson Nano is on a par with what you are seeing on the Odroid N2. With GPU (but not VPU) acceleration activated in Firefox and Chromium, I can stream Youtube videos smoothly up to 1080p30 or 720p60.

    So far as CPU performance goes, If you look at the Phoronix benchmarks for the Odroid N2, there were only one or two in which the N2 was 2x as fast as the Jetson Nano, in most of the benchmarks the speed advantage of the N2 was much less, and in several benchmarks the N2 was as little as 10% to 20% faster. Overall, I would put the average CPU speed advantage of the N2 at about 50% faster.

    On the other hand, the GPU on the Jetson Nano is not just 'a little bit faster' as you stated, because based on comparing OpenGLES glmark2-es2 scores that I am seeing on my Jetson Nano, vs those reported by cnx-software for the Odroid N2 (glmark score 2050 on my Jetson Nano, vs 1119 reported for the N2 in the cnx tests), the Nvidia Maxwell GPU in the Jetson Nano is nearly twice as powerful. It would have been nice to also compare the full OpenGL performance, but unfortunately, although the Jetson Nano has FULL OpenGL support, the N2 can't do OpenGL except useing the SLOOOOW Mesa CPU drivers (which is why your blender GUI is choppy as hell) — but since both the Nano and the N2 can do Android's dumbed down OpenGL-ES, that gives a good way to compare the GPUs.

    Since you use your SBC's heavly for blender, I would think that full OpenGL and CUDA support of the Jetson Nano would be important, since the latest Blender version still needs OpenGL for realtime rendering, and now supports dual CPU/GPU rendering. I haven't tried CUDA rendering yet on my Nano, but it should be possible since the Nano's updated Maxwell GPU cores support a fairly recent version of CUDA.

     I see that the N2 claims to support OpenCL – have you tried to get OpenCL rendering working in Blender? It might be interesting to see how GPU rendering, or mixed GPU/CPU rendering stacks up to the ARM CPU cores alone.

  11. Very helpful video. Thank you!

    Nico, I am trying to create the fastest, lowest cost ARM-based Linux HTTP web server possible at this time. Am I correct in assuming a powerful SBC such as this N2 will gain me little if anything over the XU-4 (which is doing very well)? I am running Lighttpd with Sqlite, and serve only very large high resolution photography static files. I use WordPress w/PHP 7 as well as Varnish and Cloudflare. BTW, I don't want to get into the Intel world.

    Greetings from beautiful Thailand!

  12. I would appreciate any advice. What would be the best board for a nas, considering performance and overall stability: NanoPi M4, ROCKPro64, odroid n2, or even jetson nano?

Leave a Reply

Your email address will not be published. Required fields are marked *