Orange Pi – First Fire

Well, I’ve got it working, the Orange Pi One that is.

I’ll skip the time spent wandering in the woods and get right down to the #1 issue:

It doesn’t see the HDMI monitor by default using the Armbian images.

Took me three or four iterations of “download os variation, stuff on card, try boot” to eventually just look at the log files on the chip with a Raspberry Pi and see it claimed to be booting. OK…

Then used the WiFi router (and DHCP server) to check that it had registered an IP address…

Then login via ssl and, there it is.

Just don’t expect the “desktop version” to come up with, well, a desktop…

At this point “I’m OK with that” as I’m expecting to run this module as a headless compute node, but really, if you were expecting it to work without another computer to get to it, well…

Here’s a screenshot:

Orange Pi via Raspberry Pi first login screen

Orange Pi via Raspberry Pi first login screen

Remember you can click the image to embiggen…

The Good

It is cheap. About $16 (all up a touch over $25 from Amazon with powersupply).

It is small. Slightly ambiguous in that it doesn’t fit in any of my Raspberry Pi cases, but it looks about the Altoids Mint Tin size or a bit smaller. Corners are square, though. About a thumb wide and little finger long. A stack of these would be about 2/3 the size of a similar stack of R.Pi boards with the same quad core 1.x GHz speed.

It looks like you can get a decent Armbian image or two for it. I got mine here: where, after enough clicks and things, puts you to another link from where the actual image comes.

I needed to install p7zip to unpack it so:

apt-get install p7zip

and a simple

p7zip -d {name}

unzips it. Then you just dd the image file onto the SD card and the install is done.

dd bs=4M of=/dev/sdX if={very long FOO name.img}

Where of course X in sdX is the actual drive letter assigned when you put the chip in a carrier and plug it into your Raspberry Pi…

So actual install is pretty trivial. Stick that in the O.Pi and boot. It takes a few minutes with vaguely blinking lights (red and green on the board – green is power I think… plus others near the ethernet connector) then it is up.

The Bad

The problem is you won’t know it is up. The screen very briefly gives an Armbian greeting (maybe a second) and then goes black. Nothing. Now you might think having downloaded the “desktop” version the screen would show something. You would be wrong.

A web search showed several folks and several different O.Pi types complaining about various HDMI issues and all sorts of “fiddle the config of the HDMI” to attempt to get a monitor. Some claim to have one… I am not sure I’m going to bother at this point…

It is helpful if you speak Chinese. Or at least read it. At the Orange Pi web site, the top “download” page has all sorts of nice pictures of OS types you can download and run.

Now despite many web pages saying not to trust their images as the clock speeds are too high or the overvolts are set to high or {whatever} kind of technical over-something was done, none of them were security gripes, so I was going to download an image… Except when you click on one of them you get tossed into a “pick one” page with a pretty bird on it (at least I hope that is what the page said) and clicking a link takes you to what I think is a download the actual image page… both of them in Chinese. Is it really that hard to just put a word or two in English on the page for the approximately 3 Billion people on the planet who are functional in English?

The So-So

It takes a special power dongle of 4mm? diameter, so my powersupply came as a mini-USB with an adapter (from Lovrpi). This is very nice as it includes and on/off switch. Now the board has a switch button on it too, but near as I can tell that one doesn’t do anything (that changes anything I’m running…) Their powersupply is very nice too, and I’ve used it to drive a Raspberry Pi Model 2 that needs lots of power. This is a nice solution that doesn’t commit your powersupply to the Round Connector forever.

The board, if driven to full performance, will almost certainly need a heat sink that is not included with the board. It looks like the same one for the Pi M2 or Pi M3 will fit the H3 chip in the Orange Pi. That also strongly implies that at full computes the quad core ARM at almost identical clock in the same size package will overheat without a heat sink and throttle your computes (it does that, backs off the clock when hot). So “someday” I’ll need to get a heat sink. There are web pages talking about that, too…

It doesn’t have that strange boot sequence of the R.Pi where it loads a loader to load the boot loader to load the OS… with a binary blob in the GPU running it. What it does do isn’t yet clear to me… Just “stuff and go” with the image… There are web pages saying you can compile the firmware and load it {somehow}, but many of the “how to load the OS” pages were wrong or had stale links, so not as helpful as you might want and leaves me thinking those firmware how-to pages might not be so right either. (Why there are no links to ‘how to’ pages in this article…)

It has systemD on it. Well, what do you expect from a bog-standard Debian Jessie? So along with “how do I get a monitor to work?” I’ll need to do “the usual” swap to Devuan. Which, given the results of a web search for “Devuan” and “Orange PI” had my web pages where I talk about doing it at the top… implies I’m going to be the only person on the planet running that particular OS / Hardware combination… Sigh. A “Support group and Forum” of One.

In Conclusion

I’m not unhappy with it.
I’m not happy about it.
It’s OK…

For the extra $15? I’d rather go with the Raspberry Pi and avoid the time sink. Were I making a raw compute cluster with my own OS build, it would save enough money on the hardware to be worth it (once you are into it about a dozen nodes and especially if using the even cheaper O.Pi Zero at $10… but it was not available when I made the buy)

It looks like a competent bit of hardware, compromised by a painful software download search / process / language barriers, a lack of out-of-the-zip support for HDMI, and missing heatsinks that are essential for full performance (same s Pi M2 and Pi M3…)

I do like the blue color ;-)

So at this point I’m off to see if I can figure out how to get HDMI to work, put the board on top of the Dogbone case (maybe I can wire-tie it to the top…) and use it headless for low compute load things until I can order a heatsink kit ($5 on Amazon, so need to bundle with other things or the shipping will cost more than the kit…)

Maybe tomorrow I’ll start in on that whole Devuan upgrade bit… (Or maybe just install distcc on it and start a Model E build running on a 16 core cluster ;-)

FWIW, until further notice, I’m likely to stick with the Pi Model 3 as the best combination of Computes/$ and ease of set-up with LOTS of support available. At $35 list and $50 all up with powersupply and heatsinks, if that saves me an hour I’m way ahead of the game for unit count less than 4 or 5. I may dabble with things like the O.Pi or Banana Pi, but that’s just for my own education and experience now that I have a working Raspberry Pi Cluster. Basically, it’s R&D for the question: “IFF Model E needs 88 nodes to run, that’s 22 Pi boards. Can I cut $220 off the board cost with Banana Pi or Orange Pi boards?”. So far the answer is “Yes but… at the cost of $200+ of time”…

Subscribe to feed


About E.M.Smith

A technical managerial sort interested in things from Stonehenge to computer science. My present "hot buttons' are the mythology of Climate Change and ancient metrology; but things change...
This entry was posted in Tech Bits and tagged , , , . Bookmark the permalink.

41 Responses to Orange Pi – First Fire

  1. E.M.Smith says:

    This is an interesting bit:

    CPU runs at 1.53GHz, termal management adjusted so that all 4 cores are active up to 100 °C

    One presumes “termal management” means “Thermal Management”… that is the kind of thing you run into on the ‘help’ at the Orange Pi sites…

    Now the overclock to 1.5 GHz, OK, I can accept that as it will thermal limit if that’s too fast… but they then up the thermal limit from about 70C (safe) to 100C (where dopant migration in junctions is an issues). FINE for very very short bursts of a second or two every few hours, not so good for running a build monster cluster or GCM for hours or days on end…

    I’m seeing why folks said to avoid their images… you have to back that abuse out of the settings…

    from down in the bottom:

    to always run at full speed install heatsink and fan !

    No shit Bwanna? Since I despise fans, the answer is “No!”… I’ll just run it at normal clock, thanks… and cluster them if I need more speed.

    But the good bit… looks like I’m able to “roll my own”… Likely as good as any approach…

    Building the system

    You can try to build Debian/Ubuntu for OrangePI yourself.

    Clone my GitHub repository.
    You will need running Ubuntu or Debian system (you can even run it on OrangePI).
    Before running the script install debootstrap and qemu-user-static packages.
    Read carefully and edit to adjust the parameters to your needs.
    Run sudo create_image to create Ubuntu system. I recommend to build to local directory, then you can run image_from_dir to transfer the system to sd card or image.

  2. p.g.sharrow says:

    One of the main features of the Raspberry Pi is the size of the community of users that could share the burden of porting the needed software to the available hardware abilities.
    I doubt that any cost advantage of the Orange Pi could trump that. If you want exclusivity over a possible use the blue Orange pi, it might be useful to develop this. I would fear that this might become an orphan.
    If you think migration to a better designed platform is the direction to go, that is about a 1/2 million dollar project or maybe discussions with the Pi foundation…pg

  3. E.M.Smith says:

    Small memory footprint:

    top - 18:43:45 up 2 min,  1 user,  load average: 0.12, 0.13, 0.05
    Tasks:  88 total,   1 running,  87 sleeping,   0 stopped,   0 zombie
    %Cpu(s):  0.0 us,  0.2 sy,  0.0 ni, 99.8 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st
    KiB Mem:    506132 total,    91812 used,   414320 free,     6988 buffers
    KiB Swap:   131068 total,        0 used,   131068 free.    40688 cached Mem

    So using only 91 MB out of 512 MB. One of my concerns was the small memory size, but it looks like it is compiled with that in mind (so smaller buffers and such).

    I don’t see where I need to worry about swap to SD card any time soon…

    (It now has a ‘me login’ and is sitting on top of the Dogbone Case stack of R.Pis as I look for a cable tie ;-)

  4. E.M.Smith says:


    There’s already a group devoted to tuning up the Debian port for all-things-ARM, and that is the Armbian port / group. They are relatively agnostic about hardware.

    It is the Orange Pi folks who look to be very software staff limited… some of their links just point at other parties (such as the Debian that points at Armbian) and others are all Chinese so “who knows”. One poster said he thought it looked like SW as a “one man band” at Orange Pi…

    For anything serious, at this point, it is looking like Pi Model 3 is “it”. (Or maybe an M4 if one ever replaces it…) I’ll ‘explore’ the Orange Pi, Banana Pi, and maybe even Cubie, Odroid, and C.H.I.P., but only from my personal interest in “Absolutely Minimal Computing”. Something about that topic just gets my fancy all tickled ;-) But it’s a “one week-end a quarter” kind of interest. A Sidebar that sometimes yields benefits. Like when I bought my first Raspberry Pi Model B… (still running as my DNS server, Proxy server, etc. etc…)

    Since there is a growing list of Raspberry Pi Devuan users, that, alone, makes the Raspberry Pi / Devuan combination the only reasonable choice for most folks “at this time”. It is what I’m running on 3 out of my 5 boards. (Alpine on the B as it is designed with security first and as a router platform / service board. Armbian on the O.Pi One until I can Devuan it… )

    But the very good thing is that any given class of hardware, or OS group, tends to be far more alike than different, so often tips from one works on another.. (Thus my reference to the odd Ubuntu ‘how to’ in Devuan configuring) Similarly, once in a class of OS, like Debian-Based, it’s far more the same than different. (Note that the claim is made Banana Pi images boot on the Orange Pi, and there is even a Raspbian that is supposed to work on it…) So Devuan ought to go into Armbian just as easily as it did into Raspbian. They are 99%+ identical. (Major differences tend to be administrative choices of the packager, and I end up making my own choices anyway…).

    Per “better designed platform”:

    No need, really. There are thousands of folks designing and selling them.

    It’s my opinion that the ARM chips are ‘fast enough’, and we’ll know in a half dozen staff-days of configure, compile, test of the GCMs. IFF it isn’t enough, we already know that Intel chips ARE fast enough as it is known to run on them in a cluster. “Board cost” goes up to about $100 instead of $25, but it is about 4 x to 8 x the performance too, so “Cost/compute” isn’t much different (though power cost is much higher at 200 W for typical desktop boards instead of 5 W and they are larger and have fans). The newer SBC Intel ATOM based address the power and fan issues, but at lower computes.

    Yet I already have roughly 4 Intel based machines I could cluster (after software install…) and do the benchmarking. These are old White Box PCs, and noisy power sucking things that are slow by modern standards, but as “proof of port” and benchmark, good enough. (I’d just need to adjust via known relative performance stats for them vs modern chips).

    Since in all cases I can run “Debian derived” and almost certainly Devuan, the software layer stays the same. Development / porting on the Pi works on the Atom.

    Here’s an example of the $100 “alternative” board style:

    But I’d rather not go there… even if the ATOM chip is getting away from the fan issue…

    You rapidly end up at $150 / node all up cost for board, power supply, OS load, case… for the multicore boards. Then 10 of them and you are at $1500… and that starts to be real money…

    So I’m exploring the bottom end of the pool for absolute minimal price point with acceptable computes, but so far it looks like just above the scum layer is best ;-)

    In any case, ALL the software / OS work is the same.

    Heck, you can even make heterogeneous clusters. It’s a bit more work to set up a distcc cluster that way (you need to install ‘multi-lib’ so the ARM libraries and Intel libraries are both available to the compiler tool chain…) but in fact most folks build ARM ports using cross compile from an Intel based Build Monster or Build Monster Cluster. Similar things are done in Scientific Computing. (PVM is especially suited to heterogeneous clusters). So it isn’t like you buy one arch and if it doesn’t work out you toss it. (Well, outside of government and academia ;-) It just becomes part of the total pool of computes.

    So, say, the Model E code just crushes the Pi Cluster. Takes days to get even simple results and swaps like crazy from massive memory demands. (I think it will not do that given what I’ve seen in the code, but assume it…) OK, then I start ordering $100 Intel boards and adding them to the compute pool instead of more R.Pi M3 boards. I would likely still use the M3 stack for the compiling as it is fast enough, does cross compiles fine, and is silent, but then toss it onto the Intel boards for actual model run.

    As a random example from a web search for heterogeneous MPI:

    This is MP-MPICH, Version 1.5.0, Dec 2007, an MPI-Implementation for Linux, Solaris and
    Windows NT/2000/XP/2003, supporting TCP/IP and the Scalable Coherent Interface (SCI) for
    high-performance inter-process communication.
    However, it has developed apart from MPICH, featuring an extented device interface and
    other features like:
    – MPI-2 one-sided communication,
    – support for multiple devices compiled into the library, and
    – combining multiple clusters into a big “meta computer”.
    Nevertheless, a vast part of MP-MPICH is still MPICH.

    You can even share jobs out to other clusters over the internet and they can be any odd mix of Windows, Unix, Solaris, Linux, …

    Now that was in 2007. I’m not sure how much of that has migrated into MPICH2 already. Just not “gone there” yet.

    So if my tiny Pi-Cluster doesn’t cut it, Intel bits can be glued on, and even ‘friends clusters’ over the internet can share in a giant pool….

    (There are ‘issues’ with that sharing in that the communications lag may make it better for some parts to be done locally and only some particular parts farmed out Far Far Away. Yet it is often done.)

    Well, I’m rambling….

    In summary:

    My hardware preference is only a mild preference based on fanless low power minimal cost desires; it can easily be swapped / added to without any disruption to the software layer and design. Using Intel chips / boards, or even heterogeneous clusters.

    My playing with ‘variety Pis’ doesn’t distract from the Model E cluster build / test process. They end up being used anyway, for something valuable, and sometimes contribute very useful knowledge and computes. Take today: I’ve now got the O.Pi configured as Headless3 with the .43 IP address. One or two more commands I’ll be able to run it as part of the “distcc” compile cluster “as is”. Any kernel compile now drops from 40 minutes to 30 minutes… so it will ‘pay for itself’ in time with about 6 kernel ‘makes’. I figure I’ll do that over about a 4 month period.

    At the same time, I’ve done something useful while waiting for the coffee to hit enough that I feel I can take on reading the byzantine Make structure in the Model E code ;-)

  5. p.g.sharrow says:

    It is clear to me that the ability to operate with very low power requirements is a major plus. The cost and dependability of grid power is a concern, specially with the demand that the computers must be always on and usable…pg

  6. p.g.sharrow says:

    Picked up a 6 port USB power supply;
    at Best Buy, 2+ amps on each port. Also available at Amazon $27, many others of all sizes listed, 6ft-120vac power supply cord so the block can be placed where needed …pg

  7. E.M.Smith says:

    Well that was interesting…

    I was configuring the O.Pi to run MPICH and doing my test case. Didn’t like my login. Oh Yeah, need to install “keychain”… Reboot…

    Now login remote gives “refused port 22″… so I need to start the ssh daemon again and set it to continue to restart on reboots (like it was before the keychain install…) except I can’t log in at a console because it doesn’t talk to the HDMI yet… and I can’t ssh in to fix failure to start sshd… Sigh.

    The good news is that the chip is all one big ext4 file system so I can just mount it as a file system, find the right place to put the ‘whatever’ that starts ssh again, and then boot it.

    The bad news is I have to figure out what and where…

    I think it’s just going on hold until after dinner some time… I’ve gotten it mostly all configured and ready to test MPICH (with a backup saved about 2 hours ago, so not a lot of lost work if I have to drop back to that one). But time for something more fun…


    ONE of the nice advantages of the Pi class (dinky ARM chips) is that 5 W consumption. Even found an article that says how to shut down cores when you don’t need them to get it down to something like 200mW under software control(!) no less.

    Now one big 12 VDC battery can drive a lot of 5 W loads… call it 10 W (2A x 5 V) at full load. 80 Amp Hours at 12 VDC is 960 W-hrs. Call it 1000 if you buy a slightly larger battery (to cover losses too). That’s about 200 “Pi-Hours” average or 100 Pi-hours fully loaded… 4 Pi-days, roughly. An 8 Pi cluster for about 1/2 a day. That’s 32 cores… full boat loaded.

    Charger looks interesting. The usual caveat is that phone chargers often give you the 2.4 A a PiM3 needs, or the 5 V, but not at the same time… (for the small ones). This one says:

    Advanced Charging Technology: PowerIQ and VoltageBoost combine to provide the fastest possible charge up to 2.4 amps per port

    One hopes the “Voltage Boost” is marketing lies, you want stable 5 V not boosted voltage above that… ( I’m guessing it really means “voltage sag suppression”… the usual problem with phone charger type powersupplies).

    The 60W and 6 ports is 10W if all are fully loaded, so 2A which is fine for a lightly loaded model B but 0.4 A short for a M3 at full load. So you get 4 or 5 fully burdened available.

    Not a lot of worry on using non-fully-regulated phone chargers. I’ve got 3 of mine running on them. But only one of them is a Pi Model 2 and it only works on one of the phone chargers (a nifty little ice cube like thing) and doesn’t work on the bigger 2 ports cube (where one is marked 1 A and the other 2.1 A). Well, really it will work on it, but it lights up a low volts marker on the screen sometimes. Runs, but complains about it…)

    So far the O.Pi is happy on that $couple adapter / phone charger that the M2 and M3 despise… I’m slowly buying ‘real’ powersupplies for the M2 and M3 boards. 2.4 A or better. A “gang” supply where one at a time they could pull their 2.4 at boot, or one peaks loads, then drop back closer to 2 A for normal average use would likely work well (unless you drive them all 4 cores to the wall at the same time for minutes+… which is what I intend to do ;-)

    FWIW, I’m contemplating putting a capacitor (with protection) across the powersupply when I go to a gang power supply, so surges are met by the capacitor and the average by the regulated source. First thing I ever built was a 5 VDC (tube!) power supply, so it’s kind of coming full circle for me ;-)

    Let me know how the Pi likes that ganged power brick. If it is happy with it, that would make my power strip happy and my desk way less cluttered ;-)

  8. p.g.sharrow says:

    @EMSmith; preliminary, put the Pi2 chip into the Pi3 also plugged in the passport into 1 of the Pi3 USB ports. powered the 4 Pi2s and a Pi1 as well, no indication of voltage sag on the Pi3 but none of the CPUs are actually loaded much so still no idea of how much real load this 12 amp 6 port Anker will stand.

    While powering the Pi netcruzer and Passport from the flat screen USB port I was getting the screen strike for low voltage on the Pi…pg

  9. E.M.Smith says:


    Well that’s more than enough to run what I need… I think I see a power USB Block in my future ;-)

    At that point where you are making a 12 amp feed sag from total draw, you have more load on it than is needed for a typical stack… since the way I run things I have a powered USB hub driving my disk farm, keyboard, and mouse.

  10. p.g.sharrow says:

    I would say that the Anker 12amp will drive 4 Pi3s plus a USB HHD and a benifit is that it is the just a bit smaller in size flat as a DogBone. Just set the stack on top and use 12 inch jumpers. A very clean stack. I may even glue the stack on top of the P.S. brick to stabilize it.
    Had to move the Pi3 to a DogBone as it was heating up in the snap on case.

    I would have to agree, the Pi2 was barely acquitted as a netcruzer. The Pi3 is much better for that task.
    This no fan, low power is a wonderful improvement over the Intel box and a CRT monitor. Next I need to convert a UPS to run on a large battery or 2…pg

  11. E.M.Smith says:


    Well, I managed to ‘guess’ what was the ‘download’ link on the all Chinese page at Orange Pi (later figured I’d try ‘cut / paste’ into Google translate and that worked too) and tried to download both Debian and Ubuntu from their “Baidu” provider.

    No Joy. 4 attempts (2 each) scattered around the clock. ALL failed.

    The Ubuntu didn’t even want to start. Presented the “download Ubuntu…img.xz” or whatever and choose where to save it; but then just didn’t serve any bits (so I got a zero length file of the right name…) The second attempt died partway:

    -rw------- 1 chiefio chiefio   1572864 Jan 17 07:46 ubuntu_lxde_desktop_OrangePione_v0_9_1.img.xz.part

    The Debian would download (at about modem speed…) for about 300 MB to 400 MB point then fail. Sized claimed on site is about 1 GB, but it thinks it is done at 400 MB…

    -rw-r--r-- 1 chiefio chiefio 394526720 Jan 17 08:50 debian_xfce_desktop_Orangepione_v0_9_1.img.xz
    -rw-r--r-- 1 chiefio chiefio 433586176 Jan 17 16:52 debian_xfce_desktop_Orangepione_v0_9_1.img2.xz

    Do NOT expect to download your OS from Orange Pi, IMHO… At least not for those versions. (I know “It isn’t about me” since I download GB of stuff from all over without issues.)

    I’m going to try a few more from time to time (something for the machine to do while I’m doing other things…) and see if any of them work.

    FWIW, while I’ve gotten the O.Pi to work (for headless use), the process is NOT for the novice / Noobie and certainly not pleasant even for folks who have done this a dozen times before. At this point, given the need to sink another $5 into heat sinks, the time that goes into “making it go”, and the general “here’s your sign!” attitude toward the operating system downloads, I can’t recommend it to the casual user. It just isn’t worth the $10 to $15 you save or the 1.3 vs 1.2 GHz “uplift” in speed vs the Raspberry Pi M3.

    It IS a very nice fast bit of hardware, and it DOES work as advertized and it WILL make a nice backend headless node. Heck, once fully configured and I’m making it work well, I might even order more of them (IFF I need more computes – facts not in evidence yet) since to make it go I will only need to “clone the chip and go”.

    But that points to their real strength: They are made by a hardware focused company to be sold to developers who are being paid to play with the software for days… with the intent that when they have it ‘right’ they will buy 40,000 of them to put into their IOT devices or WiFi routers or whatever gizmo they will be selling. Not the home gamer. The added costs for “unit size of one” of ordering aftermarket heat sinks, powersupply, case, SD chip, shipping etc. etc. just has that $10 cheaper being lost in the Rest Of System.

    So “it’s been fun” playing with it between “real work” tasks, and it’s now part of my computes farm, but it isn’t a better solution. Good enough, yes. For the experienced…

    Oh, and FWIW, the dinky board that I’d liked as a “WOW that’s small and would make a small stack” feature? Gets Very Warm as it is the only real heat sink on the CPU and memory chips… Surprising in a way. Never thought of the board itself as being important to system cooling (well, outside the Cray… they had copper core boards to conduct heat to a set of liquid cooled metal channels… but 750 kVA is like that, 5 W not so much). In any case, my interest in “way small” physical boards has been dampened a bit…

    The O.Pi is going to “The Farm” for a while… now on my “someday” list is to download the ‘build scripts’ for it and just “build my own” Debian / Devuan. It looks like that will be faster than trying to download their OS images…

    Per Armbian: Found a page with a statement that they shut off the monitor on purpose…

    “Coming in the 5.15 release” yet what I downloaded was the 5.24 release…:

    Prevent HDMI screen artefacts (disabling interfering TV Out by default)
    Added TV Out patch for Orange Pi PC

    already in 5.04:

    Enforce user account creation before starting the GUI

    So essentially Armbian on the Orange Pi looks to be TV monitor / GUI hostile by default “out of the can” and only lets you have the world you need to configure it after you have configured it via ssh. Sigh.

    Welcome to the world of hardware developers who don’t like a big monitor and GUI interface taking up lab bench space…


    Sounds like you are building a very nice rig!

    FWIW, the non-Raspberry Pi world has some similar quad core boards at 1.3 to 1.5 GHz (compare to the M3 at 1.2 MHz) so there is a (trivial?) small increase in speed available from some of them. IMHO likely not as valuable as a better choice of browser or an OS compiled with better memory usage in mind… or even just installing the OS to a real hard disk.

    My ‘experiments’ with various SD chips and various disks has shown me that often it is the I/O that is limiting, not the actual Pi board, and between putting swap on a Real Disk (TM) and using a class 10 card things speed up enough to notice. On my “someday” list is to compile a Devuan “from sources” and with parameters set for a faster desktop experience. But it’s a low priority “someday” so not likely this year… Frankly, the M3 is fast enough it just isn’t a real motivator to make it go faster 8-)

    Watching the “top” display, most of the times I’ve noticed sloth have been when swap is happening (so a 2 GB memory board is likely a better improvement than a faster CPU, or compile the OS to use memory more efficiently) or when an application is not ‘multi-core aware’ – like some browsers that hit 100% in one core and I’ve got 3 doing nearly nothing. Over time that will fix itself as the use of multi-core systems comes to dominate (as it is on the verge of doing now…) and developers make their code multi-threaded multi-core. (For some applications just a recompile with those flags set… already multicore in the Intel world and they had shut it off for the ARM world to reduce the multicore overhead in a single core world. But that world has changed and they have not all caught up yet…)

    BTW: I’d use a wire tie rather than glue… easier to reconfigure later when something changes or breaks… Also do a heat check on the PSupply before making it ‘bottom of stack’. Don’t want your compute cluster bathing in a heat plume… I doubt it will be very warm, but sometimes the back of the hand has ‘lit up’ over an object when I didn’t expect the warmth…

  12. p.g.sharrow says:

    “EMSmith; at present just allowing gravity to hold the stack together. No heating to be detected as of yet. Will add more load and let it run to see if there will be a problem…pg

  13. E.M.Smith says:


    I found a very nice (but complicated) script on a web site. Then proceeded to comment out most of it ;-) to make it simples… Here’s the whole thing, you can see my commented out lines plus the ‘never reached’ from the original now dominate. FWIW, the minimal set is just to look at ‘temp’ file in
    cat /sys/class/thermal/thermal_zone0/temp
    and divide by 1000… so my present reading of 56920 means 56.9 C degrees.

    So it’s possible to measure all sorts of things instead of the hand-wave over the stack (that I still do…). When doing stress tests, though, I’ll start a temperature test going…

    Basically I’m happy working in C so just bypassed all the fancy C to F conversion stuff… and I can remember that 70 C is hot and 100 C is too hot, so don’t need that nag…

    chiefio@Headend:/SG/ext/Orange$ bcat pivolts
    echo Starting Script without name hardcoded
    #!/bin/bash /root/bin/
    # The hard-coded script name here is a potential issue, but $0 fails, 
    # and we need bash to print the degree symbols
    # A script to display Raspberry Pi on-board sensor data
    # LGD: Mon Dec 23 04:07:23 PST 2013
    # LGD: Mon Aug 10 14:02:46 PDT 2015: Added degree symbols, 
    # over-temperature based on degrees C so it will work without bc
    TEMPC=$(/opt/vc/bin/vcgencmd measure_temp|awk -F "=" '{print $2}')   # Get Temp 
    #TEMPf=$(echo -e "$TEMPC" | awk -F "\'" '{print $1}' 2>/dev/null)   # Get numeri
    c-only Temp C
    #TEMPF=$(echo "scale=2;${TEMPf}*9/5+32"|bc -l)            # Calculate Temp F
    #[ `echo $TEMPC|cut -d. -f1` -gt 70 ] && ALRM=" TOO HOT! "      # Check for over
    -temp: Max = 70C or 158F
    #TEMPB4OVER=$(echo "70-${TEMPf}"|bc -l)
    #[ `type bc 2>&1 >/dev/null` ] && echo -e "Binary Calculator package (bc) not fo
    und.  Fahrenheit temperatures will not be displayed\n" # BUG: this test fails to
     work as desired
    #echo -e "\nThe BCM2835 SoC (CPU/GPU) temperature is: ${TEMPF}'F (${TEMPC}) `tpu
    t smso`$ALRM`tput rmso`"
    ##echo -e "\nThe BCM2835 SoC (CPU/GPU) temperature is: ${TEMPF}\xc2\xb0 F\t(${TE
    MPf}\xc2\xb0 C) `tput smso`$ALRM`tput rmso`"
    #echo -e "\t\t70\xc2\xb0 C HIGH-TEMP LIMIT will be reached in ${TEMPB4OVER}\xc2\
    xb0 C higher"
    echo -e "\nThe Core voltage is: \c"
    /opt/vc/bin/vcgencmd measure_volts core|awk -F "=" '{print $2}'
    echo -e "\nThe sdram Core voltage is: \c"
    /opt/vc/bin/vcgencmd measure_volts sdram_c|awk -F "=" '{print $2}'
    echo -e "\nThe sdram I/O voltage is: \c"
    /opt/vc/bin/vcgencmd measure_volts sdram_i|awk -F "=" '{print $2}'
    echo -e "\nThe sdram PHY voltage is: \c"
    /opt/vc/bin/vcgencmd measure_volts sdram_p|awk -F "=" '{print $2}'
    echo $TEMPC
    Not Reached
    On-board Hardware Sensors
       Temperatures sensors for the board itself are included as part of the raspber
    ry pi-firmware-tools package.
       The RPi offers a sensor on the BCM2835 SoC (CPU/GPU):
       /opt/vc/bin/vcgencmd measure_temp
       Alternatively, simply read from the filesystem: /sys/devices/virtual/thermal/
       cat /sys/class/thermal/thermal_zone0/temp
       For human readable output:
       awk '{printf "%3.1fM-0C\n", $1/1000}' /sys/class/thermal/thermal_zone0/temp
       Four different voltages can be monitored via /opt/vc/bin/vcgencmd as well:
    % /opt/vc/bin/vcgencmd measure_volts 
         * core for core voltage
         * sdram_c for sdram Core voltage
         * sdram_i for sdram I/O voltage
         * sdram_p for sdram PHY voltage
    The BCM2835 SoC (CPU/GPU) temperature is: temp=44.9'C
    The Core voltage is: volt=1.20V
    The sdram Core voltage is: volt=1.20V
    The sdram I/O voltage is: volt=1.20V
    The sdram PHY voltage is: volt=1.23V
    See: man sensord sensor.conf
    echo $'\xc2\xb0'C    # Works in ksh and bash
    echo -e "\0302\0260"   # Works in ksh and bash
    to enter from keyboard:
    Ctrl-Shift u + 00b0
    ��   RING ABOVE (U+02DA)   feff02da
    /root (/dev/pts/0) # echo -e "��"
    /root (/dev/pts/0) # echo -e "��"|hexdump -c
    0000000   ���   ���  \n                                                   
    /root (/dev/pts/0) # echo -e "��"|hexdump
    0000000 b0c2 000a                             
    /root (/dev/pts/0) #
    DEGCEL="\0260C"; echo -e $DEGCEL
  14. Paul Hanlon says:

    Hi Chiefio,

    Thank you for that. I was wondering about the Orange Pi and this post has settled it for me. I won’t be going that route.

    About three years ago, I bought a Cubieboard 2, and it is a great SBC, if you want a web server or NFS, otherwise probably best to leave it alone. I’m running it headless (no GUI), and except for a reboot after I updated the OS, it has worked all that time continuously. I did have one instance where a power outage corrupted the SD card, but there are ways around that, like putting the write part of the OS onto a USB stick.

    It has three things that even the PiM3 doesn’t have, Gigabit ethernet (believe me, this makes a difference, particularly for backing up and general I/O), 4GB NAND flash, so you can put a full OS (It actually comes shipped with Android pre-loaded on it), and a SATA connection for hard drives. This too makes a difference, although I never followed through on putting the second phase of the bootup process on it, with the first phase on the NAND flash as readonly, which would have been a “really nice to have”.

    I’m using it as a backup server to a 1TB Toshiba External Hard Drive, which unfortunately goes to sleep between reads and writes. I did have it connected to an internal HD, but because I didn’t “fix” the drive to something, it eventually vibrated itself to oblivion. I set up a temperature / humidity / light sensor using an Arduino with an ESP8266 (well worth your while checking this baby out), and had it sending data to the Cubie every 5 minutes.

    I’ve now garnered 250,000 datapoints into a MySQL DB on the Cubie, along with 2.5 years worth of data from the database (taken every six hours), which is very similar to the data you can get about your local area from Weather Underground. Despite network outages, neither it, or the ESP8266 (which will soon be supplanted by the ESP32) has let me down. They’ve just kept plugging away.

    I also use it as a web server, which I can access from anywhere in the world, using, which is totally free, and which has my web scripts on it. It’s blazingly fast, at least for a single user, and I’ll eventually be able to access my filesystem, do programming, get my todos, etc., etc., all from the web.

    That said, on the minus side, the GPIO pins (all 70 of them) are at a 2mm pitch, so you need special header cables to access them, cable management sucks ( the GPIO’s face downwards, the power supply for a hard drive, and HDMI interface face upwards), so forget about clustering them. And, the support for any problems one might have, is severely restricted. Definitely a board for the guru. Plus, even though they came out at the same time as the PiM1, they still cost a lot of money (about $68, last time I looked). That’s two PiM3’s.

    I’ve also tried the Banana Pi which has a similar spec to the Cubie, but cheaper (Has SATA, but no NAND, or GB ethernet). However, that died on me during a regular reboot. Also bought an Olimex, which I want to set up to replace the Cubie (for the inevitable time when that dies on me), but haven’t got around to it yet. I could expand even more, but I’ll cut it short by saying stick with the Pi. Cheap, quick, easy, have all three :-).

  15. E.M.Smith says:

    As I noted in tips, I’m presently recovering from Yet Another Power Fail (rain hard enough the satellite feed went to zero signal -measured by diagnostic on recvr – a few minutes befor the neighborhood went dark.

    This is the 3rd or 4th now, and the Pis have not cared. Just reboot on return of power. But the UPS Project just got a big schedular bump 8-{

    Thanks for the info on the cubie. While I’m settled on the PiM3 for my formal stack build, I’m going to get “variety boards” to evaluate from time to time.

    I figure I’m the right kind of guy to fight with the cranky ones, love building *Nix systems anyway, and enjoy learning the varieties of ways folks do it. Plus I can always find uses for the little guys… so time and money never wasted.

    Was reading the web pages of the NanoPi when power failed. $8 direct from China. $12 with 512 MB and good heatsink, plus they fit 8 in a 4 high dogbone case. I’ll likely order one to evaluate…once I figure out how to order from China without excess risk…

    What I’ve done so far has not needed SATA or GigE… being either memory or cpu limited, EXCEPT when doing copies of media that are IO system limited on the Pi. I figure one board with faster IO chips and I’m set.

    Did a swap partition to the Toshiba disk… then it slept and the system hung… so now I use WD or Seagate for swap and don’t buy more Toshibas…. 2 are enough… On my someday list is an OS Install to disk for Devuan. It worked well for BSD, Slackware, Gentoo, LFS, … just change an entry in the boot sector… but not while power is dodgy and my spare disk has the Mac recovery on it… it’s been a busy couple of weeks in computer land…

    Well, I need to go see why my interior WiFi isn’t talking to the ROW and find out what all still boots…

  16. p.g.sharrow says:

    @EMSmith, thanks for reminding that I need to complete my hook up to the inverter system. The system died 2 years ago. Before that it was very nice when the grid power crashed and all the electronics stayed up. It protected the electronics as well from power surges during lightning strikes and mainline wind & tree caused shorts. Now I just about have everything back in place to move the electronics line back to the inverter.

    We are enjoying the wind and rain here as well, but no outages yet from this storm…pg

  17. p.g.sharrow says:

    @EMSmith; Sorry to hear that your DNS server was taken out by the grid surge. I think maybe you need to concentrate on the needed UPS to protect your electronics. The backup feature is a plus…pg

  18. p.g.sharrow says:

    I think I will set up a secondary, Computer UPS plus a big 12v gelcell just for the electronics power strip, powered from the inverter and main bank. WE live on the end of a long grid line and the voltage surge we get is sometimes massive. One surge 2 years ago from a tree caused short, caused a flashover at the pole transformer, burned half the paint off of the pot! The neighbor lost his electronics, no problem to mine but it did damage the UPS charger…pg

  19. p.g.sharrow says:

    My grandson is setting up a Intel-windoz box for me.
    Optplex 780 cpu, 8 core, max. at 48 gigaflops. ( he seems to think that this is special ;-))
    Guess I will be able to use Acad again!
    Hopefully we will have a small 3D printer up in a few weeks, and I will need to create a new small size fume scrubber…pg

  20. E.M.Smith says:

    There comes a time when you need to think “Maybe it isn’t about me” and then comes, often much later, “Maybe it isn’t the software either”… and you ask yourself “Could it be the hardware?”. Admin / Software types typically have a strong barrier at that point. They assume the hardware is good. 99.9% of the time that’s the correct path, so why question it.

    But eventually “That day comes”…

    That discussion looks at the HDMI failures of Orange Pi boards. It is from 2016, so about older Orange Pi variants, but designers typically carry forward most of their basic design choices into subsequent generations of boards; concentrating on minor changes like adding or removing particular connectors, or changing memory size.

    So what’s in that thread? I’m going to preserve most of it here for the simple reason that it basically says the hardware for the HDMI port is fundamentally flawed and WILL fail under minor electrical stress. Since I’m in a low humidity climate, and since I “plug and unplug” the HDMI connector often, it is a near certainty I’ll blow out the HDMI interface in short order in my typical use profiles.

    Post time 2016-5-21 08:54:55
    It’s been 3 weeks since my OrangePi PC stopped showing video from HDMI. I didn’t want to touch anything since I’m presenting a project next week and the board is working without any problems through SSH. Everything works but HDMI, and I don’t know what happened, just one day stopped showing when I turned it on in the morning and never came again.
    Did someone experienced this and know if there’s a trick to solve it? I tried different HDMI cables (I always been using HDMI-HDMI) and I even tried it on different TV’s but no avail, it is not sending video signal. The port is physically in a good shape, so it should not be that problem I think.

    I also installed a clean Armbian 5.10 in other microsd and same, but I could enter through SSH too. It’s sad because I wanted to do some graphical tasks once I finished my project but it seems that I will have hard time with it . So pity I just bought it three months ago.

    Any suggestions?

    Thank you very much for your help.

    Post time 2016-5-22 03:36:20
    I’m in the same situation. I think the problem can be design, this board does not have any protection in the HDMI.

    This board has no ESD protection on bus HDMI output, and this makes it burns easily.

    NO electrostatic discharge protection on the HDMI output bus. So for the saving of the cost of a single part, maybe 1/4 of a ¢, the entire video out is at risk from normal conditions. Sigh. There is a point where “cheap is too cheap”, and this is waaay past that.

    Post time 2016-5-22 03:50:06
    Yes, missing ESD protection on HDMI is a problem. My friend also destroyed HDMI port on his OPiPlus, although TV out still works.
    Post time 2016-5-25 07:50:50
    I see. Isn’t any way to fix the part that have burnt? it’s nonsense to have the computer like this, and now more since I’ve finished the project I had and want to do some graphical stuff.

    If there is no solution I will buy an Odroid C2 which seems to be way better, I think it would worth the extra money.
    Post time 2016-5-25 23:24:59 | Show the author posts only
    hdmi connector goes directly to the H3, if something is damaged it is in the H3

    So you have a lead-out directly from THE main chip that is the System On A Chip, that has zero protection from ESD, and it taken to a large connector where it is likely to be picked up, touched, connected and unconnected and generally exposed to charge flux.


    “What were you thinking!?” comes to mind…

    Post time 2016-7-7 20:14:49
    However, you can get a serial console with a 3-pin header and any USB-TTL dongle.
    Post time 2016-8-8 17:21:10
    Edited by DeeJayOne at 2016-8-8 18:10

    I have had some problems too with the HDMI outputs.
    I’m using an external HDMI switch and after connected it the Orange Pi PC about 2-3 times, the HDMI switch was broken.

    It also seems that my HDMI switch doesn’t have any ESD protection component.
    So if both the switch and the device (Orange Pi PC) don’t have any ESD protection…

    I don’t know if there is a new hardware rev. for this board which include this ESD protection components. If not, it could be nice to do it.
    Rem. : My Orange Pi PC rev. board is v1.2

    The Orange Pi PC Plus seems having this ESD protection, so I can use it instead without any worry.
    The Orange Pi One have it too.


    Now this poster asserts that ESD protection has been added to the Pi One, so this ought not to be the issue with my board… but still… There’s clearly a low priority put on getting HDMI right and protection for parts of the chipset. (bolding by me, BTW)

    Post time 2016-8-22 16:59:34
    Lol wtf.
    Meaning I just bought a “deliberately flawed” board.

    So my overall conclusion is that the Orange Pi family has “way too cheap” hardware design choices (weak and exposed HDMI design, no heat sink when normal use profiles require one, memory vs CPU vs I/O violation of Amdahl’s Other Law (see: ) along with poor software support.

    Oh Well… It costs about the same as 2 Starbucks Vente Mocha with whip… and has give at least that much fun in the exploration. Plus, it still makes a fine headless node (once I get the heatsinks installed).

    FWIW, this came about after I installed “DietPi” on it and had the same problems. At that point, with three different OSs all having the same video failure (and the complaints on the DietPi forum being about NOT having serial output enabled by default but only HDMI…) it was strongly pointing to “Not me (config / install process) and Not OS / Software”. Now it still could be me, but at this point the ‘value added’ from finding that missing Secret Sauce is strongly diminishing (if it exists at all…) and I also know I’m not buying more of these things so that time spent will NOT be amortized over a large buy / install process later.

    At this point I’m just going to explore converting the Debian based OS on it (headless node packages) into a Devuan just to see if it can be done. (It ought to be a ‘one line change’ to the /etc/apt/sources.list file, then an apt-get update ; apt-get upgrade). After that, it just becomes an ignored board out in the Compute Farm that takes distcc runs from time to time…

  21. E.M.Smith says:


    Oh Man! What are you going to use 8 cores of Gigaflops for? (I’m so jealous … ;-)

    FWIW, “My old Cray” was $40 Million and did 400 MEGA-flops… so your 48 GIGA-flops is “only” about 100 times more compute power…. Just sayin’…

    The DNS server was only a config issue on my part. I’d removed a USB stick from it and removed the mount from the /etc/fstab file, BUT, I forgot to do the “lbu ci” to save those changes… so when it came back up from the powerfail, hung at that ‘mount this filesystem’ point… Being headless, I couldn’t see that until I plugged in a monitor… So it was a simple fix. Comment out line in /etc/fstab (again…) and “lbu commit” to save it, reboot. Now it’s all fine again.

    Yes, I’ve been “reminded” too that our power grid is returning to UNSTABLE now that Democrat Gov. Moonbeam has been in power long enough to put is in “Democrat Mode” again. (BTW, not a pejorative… just a correct description of events. Grid goes unstable under Democrats – worst was Gov. Gray “out” Davis – and became stable again under Republican RINO Aaahhhnold…)

    Plus the “gusty winds and blustery storms” of the ’50s and ’60s have returned, so more wind caused outages happening (gee… like I remember from back then ;-)

    So I’ve now got the Pi Stack and my DNS server and desktop on a $40 UPS from Walmart (not big enough for more than about 10 minutes with the monitor on it, but good enough for a controlled shutdown until the rest gets done… or will go for a few hours with the monitor off – i.e. when I’m away from home on errands) and a battery on the way for a 250 W / hour or two UPS fix for the “rest of the office” (and into which that $40 job will be plugged… so stack goes last).

    BTW, that chain of small UPS to Large UPS goes to “surge suppressor power strip” before reaching the wall… that is a GFI circuit breaker outlet… It is highly unlikely any surge can make it from wall to even the second UPS and certainly not the computer stack. On my “someday list” is to buy a couple of surge suppressor / GFI breakers for the panel out back… but just hasn’t been needed enough to do yet.

    You might consider one of these:

    They are used in hospitals (for the “hospital grade” that will mostly be the inspection / certification process IMHO) and are a nice easy wall upgrade for anywhere you plug in sensitive gear.

    Then in the next week or three I need to get that kW UPS off the bone pile, that died when the batteries gave out. Needs a refurbish and battery replacement and goes back in the Living Room. It had been driving a nice floor lamp and the entertainment cluster in prior years; so in a power outage (under Gray “out” Davis they were common) would not interrupt the evening unless it lasted longer than an hour. It was nice to “wonder what that cheeping noise is about” while watching the evening news and have to go looking to realize the power was out 8-)

    That, and I need to do a ‘service’ cycle on the Honda Generator. Have not run it in 3 years and I think the fuel system needs cleaning… it doesn’t want to start (but was always a bit hard to start when not run for a while).

    I think I also need to finally get a Round Tuit and put a 100 AmpHour RV deep discharge battery into the battery box, attache a ‘car charger’, and bolt on my large inverter… Gee, just were I left off when we dumped Gray “out” Davis… That to run the ‘fridge and a few more house lights without starting up the generator. Ought to get us through most outages. Only two in the last 30 years has been longer than the “couple of hours” that would give us.

    By this time next week I’ll have the office done, and in a month the living room. Kitchen “someday”…

  22. Larry Ledwick says:

    Not a real solution but a crutch to help keep them alive perhaps. Unfortunately it costs about as much as the computer and interface it would be protecting.

  23. Pingback: DietPi – a Small Pi Linux | Musings from the Chiefio

  24. Pingback: Odroid-C2 Performance and Order | Musings from the Chiefio

  25. E.M.Smith says:

    Well, I’ve tried a bunch of stuff, including different OSs, and I think the hardware is NOT fried from ESD. On a fresh new install of Armbian, it puts up the start of the console record, then once again goes blank. Not “No Signal”, but blank.

    IMHO they are presuming a particularly high end monitor with fast scan and lots of pixels and I’m on an old ViewSonic with none of that, and it just blanks. OK, I’d have presumed a standard 2001 HDTV resolution, but whatever…

    I tried fiddling the resolution bits at boot time, but need to do more to figure out what uboot is doing. It may end up being a fundamental incompatibility of monitor to HDMI software (but is more likely some config issue). Not really caring about it enough anymore, I’ve decided to just put it to work…

    Oh, and for Armbian, only the “legacy” kernel supports HDMI monitors… the mainline doesn’t, so you MUST install the “Ubuntu Desktop” to have it work… Sigh.

    I’ve put the Ubuntu Desktop image on it (problem didn’t go away though) and did a basic config (my login, nfs-kernel-server, etc.). It is now up and running as a data gathering appliance.

    Essentially, I’ve got it downloading temperature data from a couple of places and stuffing it into a wget archive on a USB disk. Since that task needs doing on an ongoing basis, I can just set up a cron job to keep that data up to date.

    Then, it serves that data via NFS to the rest of my machines.

    I’m basically just going to leave it doing that job for however long it takes for me to get interested in it again. For now, at least, it’s got work to do…

    At some future time, I’ll also give it the job of making my Debian / Devuan Source Mirror and serving that up too.

    It is reasonably well suited to this kind of task. Doing nothing for hours on end, then serving a file for a few minutes, or fetching one, then cooling down again. 4 cores means it can be doing 2 fetches and 2 serves without anything leaving a running core, and that’s more than I expect at any one time.

    In any case, it is something I’d wanted to make “someday”, and now this little card is doing it. Kind of nice, in a way, to look over and see the ethernet light blinking every so often and know it’s fetching the latest temperature data update…

  26. E.M.Smith says:

    FWIW, the Orange Pi as a headless net-scraper and nfs file server is doing nicely.

    top - 23:13:49 up  7:25,  2 users,  load average: 0.04, 0.20, 0.13
    Tasks: 139 total,   1 running, 138 sleeping,   0 stopped,   0 zombie
    %Cpu(s):  0.5 us,  0.2 sy,  0.0 ni, 99.2 id,  0.0 wa,  0.0 hi,  0.2 si,  0.0 
    KiB Mem :   506132 total,    87572 free,   118192 used,   300368 buff/cache
    KiB Swap:  2228216 total,  2228216 free,        0 used.   362967 avail Mem 
      PID USER      PR  NI    VIRT    RES    SHR S  %CPU %MEM     TIME+ COMMAND  
     6535 chiefio   20   0    9124   4324   1128 S   1.3  0.9  14:00.53 wget     
     6701 chiefio   20   0    6564   1432    944 R   1.3  0.3   4:41.80 top      
        3 root      20   0       0      0      0 S   0.3  0.0   0:35.11 ksoftir+ 
       24 root      20   0       0      0      0 S   0.3  0.0   0:59.70 kworker

    Nearly no CPU being used at 99.2% idle, but with an active site scrape running via wget, the ‘top’ showing status in one window, and scrolling output of the wget in another window on my Daily Driver Desktop. Waaayyy over kill on CPU power, so not going to get anywhere near that 25% to 50% ‘getting warm’ workload.

    NFS mounted to my DDriver and I can look at the files, whatever, without trouble.

    This also frees up the DDriver and the ‘distcc’ or ‘gcm’ stack for other tasks, software updates, chip swaps, “whatever” without worry about what is nfs mounted from where or do I have a site scrape running.

    It is nice to be able to reboot at will, swap OS or chips, and just not worry about the background jobs and tasks.

    So count me as a “happy camper” on the Orange Pi Buy as an infrastructure headless server. For $16 bucks, not too bad ;-)

    I’ll still worry it a bit about how to get my HDMI screen to display, but only on the odd slow day when I’m not doing anything else. It would be nice to stress test it as a full on desktop and see how it does, but clearly that’s just “chip abuse” for no real benefit. It needs a heat sink and some parameter tuning before that makes sense…

    And it needs to finish updating the climate data scrape… looks like the last time I did it was about a year+ ago… it is gathering a lot of 2016 files that are not already in the archive…

    So there you have it. A happy note after all the grousing about HDMI…

  27. E.M.Smith says:

    Golly…. Took the speed brakes off of the wget command doing the data scrape and realized just how fast my network is now compared to when I first wrote the script…

    CPU jumped up to between 10% and occasionally 25% used.

    top - 07:30:58 up 15:42,  2 users,  load average: 0.50, 0.39, 0.34
    Tasks: 139 total,   2 running, 137 sleeping,   0 stopped,   0 zombie
    %Cpu(s):  1.0 us,  6.2 sy,  0.0 ni, 91.7 id,  0.4 wa,  0.0 hi,  0.6 si,  0.0 
    KiB Mem :   506132 total,   107580 free,   116800 used,   281752 buff/cache
    KiB Swap:  2228216 total,  2228216 free,        0 used.   363623 avail Mem 
    add filter #1 (ignoring case) as: [!]FLD?VAL 
      PID USER      PR  NI    VIRT    RES    SHR S  %CPU %MEM     TIME+ COMMAND  
    11588 chiefio   20   0    6228   1116    812 R  24.8  0.2   2:19.07 wget     
        3 root      20   0       0      0      0 S   2.6  0.0   0:59.03 ksoftir+ 
       69 root      rt   0       0      0      0 S   2.0  0.0   3:07.61 cfinter+ 
       38 root      20   0       0      0      0 S   1.3  0.0   0:43.81 kswapd0 

    So a reasonable workload for the O.Pi given that heating issues start at about the 25% utilization level without a heat sink.

    I’d originally set the speed limit at about 1/2 my {then} bandwidth. Now that’s about 1/30 … and removing that limit has let things go a whole lot faster…

  28. E.M.Smith says:

    Amusing parallel processing at the moment…

    I bought a new 4 TB disk to hold temperature and related archives. One board (Pi M3) is busy for the next 20+ hours moving that data from other disks onto the new one (along with some systems deep archives). All USB 2.0 and I/O speed limited. So that system is busy for the next day or two.

    At the same time, the Orange Pi One has been running a few days (and likely for the next few too) downloading new historical record site scrapes from some temperature sites. So that card tied up for some days, speed limited by internet speed and I/O on a different network.

    Then I’m running my desktop on a third…

    No, none of it really using many cores, so much idle CPU, yet the systems are fully booked on those jobs. I don’t know of any nifty acronym like MIMD that describes this kind of parallel. Heck, even COW isn’t quite right as that still implies similar work.

    Well, back to work…

    Oh, but first: Thanks to donors for the giant disk filling with temperature data and related!

  29. Pingback: Scraping NOAA and CDIAC | Musings from the Chiefio

  30. E.M.Smith says:

    success at last…

    From here:

    The “special sauce” is the “h3disp” command. You must feed it a “-d” option to use an HDMI to DVI adapter cable… I got it working with:

    h3disp -m 33 -d

    Now I need to find out what mode I really ought to be using as this one isn’t as much resolution as I usually get… OK, but not great. But hey, I’m typing this in an X Window session on the Orange Pi on Armbian! Success! At last…

    I’m likely to use it as my Daily Driver for a few days just to get a feel for what it has in the way of “issues” now that I can get to things. (Firefox works, so that’s a “plus 1” over the Odroid 64 bit build…)

    This will also let me more easily do those comparitive heat vs load tests on it. Then add the heatsinks (that have arrived) and try again.

    OK, enough of this, now I’m going to shut it down and move cables and bits… and set up to run it “headful” for a while and not dangling on a wire off of my hard disk cabinet ;-)

  31. E.M.Smith says:

    Ah, looks like mode 36 gives me the right resolution along with round circles, so:

    h3disp -m 36 -d

    For want of a nail…

    Note To Self: Armbian is “different” in video handling. It has a ‘special’ program and it doesn’t auto-detect HDMI to DVI. You must boot the Orange Pi, ssh into it from another system, then become root, then run “h3disp -m 36 -d” or whatever mode you like. Running “h3disp” with no arguments gives you a list of choices.

    Sigh. I do wish folks could standardise on how to do system admin in Linux Land.

  32. E.M.Smith says:

    Well… That temperature test may be quicker than I thought…

    Running “armbianmonitor -m” to get temperature and CPU usage.

    At “running a desktop with ‘top’ in a terminal window”, it is already at 62 C (1008 mHz) then drops down to 480 mHz but stays 62 C. Do much of anything, like launch Firefox, and it jumps up to 1200 mHz and cruises up to 76 C. It then hangs around “70 something” while using the browser.

    As IMHO 70+C is already hot and would benefit from a heat sink, clearly doing anything that uses more than about 1/3 to 1/2 the processor power is going to put temperature through the roof. Typing this I’m about 1/3 the CPU power consumed and 73 C. (29% cpu per the report)

    Also there are delays on video things. So grab the window bar of the browser and wiggle it around, you get ‘tearing’ artifacts galore, that then take a long time to go away while firefox and something called “web content” are strongly D I/O waiting in the “top” window.

    It’s just noticeably a bit slow when you do things. Not at all what would be expected from a quad core CPU at 1200 mHz. So something isn’t quite right in tuning land.

    Given this “slightly doggy” feel and all, I’m unlikely to stay on this as my Daily Driver for long.

    On the good side, FireFox is working as advertised and I’ve got lots of desktop real-estate with very high resolution now…

    OK, I don’t know if it is an xfce (vs my usual lxde) desktop software issue, that memory is 1/2 as large issue, or that it’s just a young port of Linux NOT tuned to graphical desktops but tuned to IOT developers. Whatever. It isn’t the desktop experience folks would expect and is significantly less pleasant that the Pi Model 3 (about like the Pi Model 2 old version…) I wonder if they just didn’t bother to use the GPU since the target market is embedded things?…

    With that, I think I’ll do just a bit more heat testing and then put on the heat sinks and re-measure. I want that done so I can go back to my regular system ;-)

    Gee, added a dictionary and spell checked the edit on this comment and temperature jumped to 75 C… That at about all of 30% to 35% of CPU…

    Clearly way under cooled…

  33. E.M.Smith says:

    OK, I did a set of stress tests with no heat sinks, then added them and repeated. These are just the little stick on ones used with the Raspberry Pi.

    First off, now that they are on, typing this is not showing the typeahead and lags of before. I think I know why. Notice that “75 C” at the end of the comment above? That’s the point where it thermal limits and starts stealing cycles and slowing the clock rate.

    Now at a full 1200 mHz in the browser I’m running just 64C to 66C and it isn’t backing off the CPU speed. (Other than when there is nothing to do…)

    The test was trivial. Open a terminal window, in it, put an infinite loop stuffing a variable with a value:

    while true

    Do that in three more windows, one at a time a few minutes between each launch. (The last one I made the assignment statement: TEMP=”Sucky” )

    No heatsinks:

    1 core 100% CPU per TOP (yet only 7% per armbianmonitor -m, go figure):
    1200 MHz at 69 C at launch. Seconds later:
    912 MHz at 75 C, so already temp limited and scaling back. temp dropped to 72C

    2 cores at 200% CPU per TOP:
    1200 MHz to 75 C
    912 MHz at 78C

    3 cores at 300% CPU per TOP
    912 MHz at 78C at launch (see above 2 core end state)
    then to 75 C then back up (likely just a thermal oscillation artifact from the drop to 912 MHz and my being impatient about launching the 3rd core…)

    4 cores at 400% CPU per TOP (though oddly 25% per armbianmonitor… go figure)
    912 MHz to 78 C
    768 MHz to 80 C
    and I rapidly killed the test. Likely ought to have let it run longer as later we find temperature clock oscillations.

    Shut down, added heat sinks. Reran the test:

    Started at 51 C at 7% CPU doing prep for the run. 1008 MHz to 480 MHz at times.

    1 Core: 62 C 1200 MHz
    2 Core: 71 C 1200 MHz
    3 Core: 75 C 1200 MHz “for a while”, then 912 MHz 75 C to 1200 MHz to 912 MHz…
    4 Core: 78 C at 912 HMz dropping to 72 C then back up to 1200 MHz to 79 C to 912 MHz and so on and so forth with about a 6 second oscillation period.

    IMHO without heat sinks you have a nice little single core board that can almost run a browser but overheats…

    With heat sinks, you can use 2 cores full on, but 3 is a stretch. At 4 you are being limited back to 2.5 …

    In all cases, the “Web Content” program (system D related?) causes browsing and GUI to the Internet stuff to be painful as it “D” wait states.

  34. E.M.Smith says:

    And with that, it’s back to being a Web Scraper and NFS server headless node. Nice to know I can make it headful if I want it to be, but a PITA as a Daily Driver due to the software being a bit daft about things GUI and Internet. (Don’t know if that’s an Armbian thing, or an Orange Firmware thing, or systemD isms or what, but it’s not gonna cut it as a desktop.)

    I’m happy just knowing that it’s a 2 core server (with 2 to 4 second turbo to 4 cores before it thermal limits back toward 2…) and use it accordingly. For $16, it’s a fine one of those.

    BUT, it simply can not be a decent compute node in a cluster where CPUs are driven to the wall with 100% load on all cores for hours on end.

    Further, since it’s an H3 chip, we know that ALL H3 systems will be that way unless there is a big honking chunk of metal bolted onto them. The “stick on” gives a nice boost, but not enough.

    Also, by extension, knowing that the A53 and A57 64 bit cores burn even more energy, we know that any systems with them in them will also need big heat sinks… Like that Odroid-C2…

    Which then further points out you can get a meta-understanding of the makers:

    NanoPI and Orange are happy to ship systems touted as 4 cores where you can only use one.

    Odroid ships their system with a well balanced heat sink already installed so you can actually use it.

    So from here on out, my first stop for more boards will be either the Raspberry Pi M3 with aftermarket heat sinks and / or the Odroid family, depending on what is needed.

    Dinky boards, those with no heat sinks and inadequate cooling, and systems that thermal limit due to being built on tiny little cards hold little interest, despite the cheap price.

    With that, I think I’m done with the Orange Pi One testing, shakedown, configuration and all. It can now go off to the DMZ as a nice little worker bot machine doing useful work with a couple of cores and a small disk…

  35. p.g.sharrow says:

    @EMSmith; mission accomplished. The RasPi-3B seems to be a good stick to measure the others against. The Odroids seem to be a better balance of IO speeds but that huge sink indicates too much computation density.. The lower density and heat dissipation needed of the RasPi’s might be a feature. I’ve been examining the DogBone stack of RasPi’s that I have and they appear to me to be poorly spaced for maximum dissipation under natural convection conditions. A DogBone stack of RasPi’-3Bs may well be a bridge to far under long full use. Just suspending the unit on it’s side, GIPO dip down would help. Doubling the space of the backside to plate would also help. The little stick on sinks for the Pi-3B seem to be just enough help…..maybe
    I’ll see if I can create a better nesting arrangment for the Pi form factor…pg.

  36. E.M.Smith says:


    I’ve got my heat sinks stuck on so that they align side to side. Next up is a heat test of them in the dogbone stack, but the reason I have the h.s. side to side is so that when the stack is laying on the side (GPIO pins down) convection ought to be maximized…

    It would be trivial to make the spacing larger, just get longer spacer rods with screw ends. They are a generic part and you can order them by the dozen any length you like. I’m not seeing the need though, in a sideways stack at least. I’m a bit worried that “normal way” the top board will be in a heat plume from below, but so far I don’t have a use where that’s become a problem. As usual “we’ll see”.

    FWIW air flow in a plenum in non-intuitive at times. Spaced too far, counter currents and circular eddies try to form. Spaced too close, not enough air channel. “Just Right” you get a warm plenum / chimney with maximal draw. I suspect that a side mount is near maximal draw… Even in the upright position, it is possible that air flow would naturally flow in one side (likely the far end…) and out the others (GPIO and HDMI sides) in the normal orientation. The heat plume from the chip rising to the acrylic layer above and forming a flat flow sideways / out while cold flows in low down. and from the colder end.

    Maybe I’ll fire up the whole stack with a test case and see where it feels warm… I could even use birthday candle smoke to visualize air flow… (or set my desk on fire ;-)

    Or just fire up my trivial test case and report the “pivolts” program temperature readings and call it done… but where’s the fun in that? No fire. No explosions. No toxic gasses ;-)


    But at least this shows a good narrative of how an exploration happens. Starting with enthusiasm for a “quad core damn fast cheap board” but with theoretical concerns. Seeing the concerns actualized and others show up. Beat back the “issues” while measuring to explore the concerns. Ending at a better understanding. Discovering unexpected understandings.

    Small boards are heat problems.
    Anything bigger than one core needs a heat sink.
    Cheap is often TOO cheap.
    Software matters rather a lot more than expected.
    Community size drives software quality.
    There is a lot more to real world performance than MHz and word size…
    It is worth paying up for better design work.
    Aluminum is the cheapest computes you can buy.
    and on and on

    Then in the end matching the device to a job “just right” for its size shoes…


  37. p.g.sharrow says:

    yep, screws and spacers ordered, to move the board 1/4 out to 1/2, more then that can result in eddy currents, so should be about right for convection on the back side, the 1/4 inch might have been fine without the crap that restricts that space down to 1/8th inch. I agree that fans suck! almost as much as hot spots;-)…pg

  38. p.g.sharrow says:

    Oh yes, Will repackage some and send them along…pg

  39. E.M.Smith says:

    Just did the Raspberry Pi M3 in the dogbone stack. Stayed at 1.2 GHz the whole time. Temps taken at 20 second intervals. Initial temp 51.5 C

    1 core: 54.2 55.8 56.9 58.0 59.1
    2 cores: 62.3 63.4 64.5 65.5 66.1 62.1 stable after that at 62.1
    3 cores: 70.9 72.5 74.1 75.2 75.2 77.4
    4 cores: 78.4 80.6 81.1 81.7 81.7 82.2 repeats 82.2 after that.

    So it looks like for any given level of computes, the Pi M3 holds the frequency stable and lets the temperature rise to a slightly higher reaction point of somewhere about 80 C where the Orange Pi cuts juice at about 75 C (that it starts to bump at 3 cores, like the Pi)

    I could sense no perceptible warm air around the periphery of the dogbone stack, so by the time it dilutes off the heat sink the temperature of the air is quite low.

    A single box fan of low speed at one edge of the stack would cool all of it to below 80 C with ease, if needed, but few parallel tasks can peg all 4 cores at a consistent 100% load for minutes on end. The communications lag / time sees to that generally, and the problem decomposition into separate tasks is never that perfect…

    Oh, and that was in the regular orientation of the dogbone stack. Not on it’s side like the Orange Pi was. Hmmm… Hang one a mo, I’m going to tip the stack and re-run.

    Gee, about 2 C cooler.

    2 cores: 60.0 62.8 63.4 63.9 65.5
    3 cores: 68.2 69.8 72.0 72.5 73.6 74.1 repeats
    4 cores: 77.4 78.4 79.5 80.6 80.6 repeats

    To show core clock frequency do, as root:
    cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_cur_freq

    I could have done cpu1 and etc, but as they share a die figured it would not be that different…

    Since I’m comfortable with 80 C and the other three boards are 2 x Pi M2 (so less heat) and the Odroid with lots of metal, I think I’m fine to just “stand” it on the side, GPIO pins down.

    BTW, found a relative performance page on computes vs freq for ARM chips. Posting “soon”-ish.

  40. E.M.Smith says:

    Well, on full 100% x 4 cores (or really 400% plus browser and more) for a few minutes, the Odroid not only is letting me type fast and easy (using Ubuntu MATE no less, not exactly a slim efficient OS) but it isn’t getting all that hot, either:

    Hostname: odroid64
    CPU Frequency: 1536Mhz
    TEMP: 67

    then after a few minutes

    Hostname: odroid64
    CPU Frequency: 1536Mhz
    TEMP: 71

    And a bit later after loading and typing all this:

    Hostname: odroid64
    CPU Frequency: 1536Mhz
    TEMP: 73

    So “takes a punch” very very well. I’ll cross post this back where I did the other heat /speed tests too. But for a higher end compute node, the Odroid-C2 has it. (Though I still need to get Devuan onto it… at present it is Ubuntu with SystemD).

  41. E.M.Smith says:

    Oh, and just to finish squaring the circle… Once you get the GUI and HDMI working, you find Armbian launches a full desktop with autologin active by default. If you DON’T want that on your card after all this (and sending it to The Farm as a headless server) you can get back to a Login: prompt via a systemd-ism (as it is running Ubuntu):

    Posted 01 August 2016 – 07:43 AM

    I came across this topic while searching for the same answer.

    First you want to disable the display manager service, but what’s it called? Find out what services are display related:

    Here’s how I found out which service to disable:

    chris@trippy:~$ systemctl list-units | grep -i display
    nodm.service loaded active running LSB: No Display Manager

    Now, I want to disable the service called “nodm” (which is a cute but very confusing name):

    sudo systemctl disable nodm
    sudo reboot

    So the key bit is that “systemctl disable nodm” …

    One presumes to resume desktop / autologin one just does: systemctl enable nodm

Comments are closed.