Have you retried with a more recent image? After I kicked them about the USB Ethernet situation Ndec enabled a few more drivers for the kernel in the Ubuntu images. Both my ancient Pegasus USB1.1 and a newer USB3 GBit card then worked out of the box.

The WiFi situation is indeed annoying and I've complained about it, they seem to be since looking into it, but as the driver is open source, someone else could check too and send patches. There even is a debug tool to intercept the proprietary QC driver commands (used in Android images). It should just be reconfiguring the filters away from their broken default state.

Yeah, I had a similar situation on the Hikey 96board. I ended up downloading their kernel source and building a bunch of additional modules so I could get USB Ethernet, as the wifi was basically useless there too.

Yeah, luckily that was very straight forward. Did that too before they released the new images.

https://bugs.96boards.org/show_bug.cgi?id=66 "Ubuntu can't ssh in"

[Slightly offtopic] Since you seem to have lots of experience (and numerous ARM boards), I'm curious if you've had a chance to try Applied Micro's 64-bit ARM SOC[0]? I realize it's $1500, but that's probably to reduce support costs and qualify potential customers. Anyway, it would be nice to know from a first-hand source.

[0]https://www.apm.com/products/data-center/x-gene-family/x-c1-...

We've been using these at Red Hat for 18 months+ and after an initial rough start, they are now our workhorse machines for aarch64 development (I've got one under my desk). They are 8 core (plus 1 x 32 bit management core), come with 8, 16 or 32 GB of RAM, UEFI firmware, ACPI and other enterprise features. They're nice and fast for development work.

Largely, the hardware and firmware bugs have now all been resolved or worked around.

Downsides: No wake-on-LAN. No IPMI. (Both rumoured to exist in the management core, but I've not worked out how to enable either). The 10 Gb ethernet doesn't work for me. I don't have a monitor plugged into mine, but that's OK since I only care about serial access for ARM server hardware. Mine is noisy because of the numerous, loud fans. I disabled all the case fans, but haven't had the balls to disable the processor fan which still makes a racket.

For virt development, OpenStack and that kind of thing, they're the only [aarch64] game in town.

In summary, they work well.

Thanks for the run-down, esp. the mention of the CPU fan, as I had noticed it in one of the photos. I'd be interested in reducing clock speed in the hope that I could get rid of all the fans (maybe use a heatsink?). The available RAM and lack of GPU are major benefits, and I hope we see more products with those features. Cavium seem like they could offer something similar, but I don't think their sales model fits with widespread availability of dev kits.

Edit: Their site and sales sheet have a discrepancy about power supply. Does the dev kit come with one, and what are the specs? Thanks again!

There's a photo of mine on this page (on the left): https://rwmj.wordpress.com/2014/09/08/raise-the-itanic/

The machine is physically a standard 19" 1U rack case, half depth (or less?). It comes with an 80W PSU and a single SATA HDD. I fitted mine with a second SDD (I believe there are 4 SATA ports on the motherboard, but not enough room in the case for 4 hard drives).

The motherboard inside is pretty tiny, about 7"x7".

Red Hat have some Caviums and Seattles as well, but I've not used those.

Edit: I forgot an important point: The upstream kernel works out of the box. (For x86 users, this will sound like a strange thing to point out, but it's BIG news in the ARM world).

What is the power draw like at idle and load?

I only have a kill-a-watt for measuring power consumption. I am using a brandless Universal AC Adapter at 15V and I do not know its efficiency. I'm not aware of any ability to monitor power consumption directly on the dragonboard.

    Idle, text  1W 5VA
    Idle, X     2W 5VA 
    1CPU, X     3W 7VA
    2CPU, X     4W 9VA
    4CPU, X     4W 11VA

(CPU load generated by dieharder statistical randomness test on AES_OFB CRNG, dieharder -a -g 205)

Incidentally, Dieharder's AES_OFB CRNG (which I don't think uses the AES instruction set on intel) runs at 5e6 randoms per second per thread on dragonboard, compared to 2.5e7 randoms per second per thread on my 3yo 2-core/4-thread i5 laptop at 2.6GHz, to make just one simplistic comparison of CPU speed.

(edited to try to fix formatting)

The board should be prepared for current measurement instrumentation. Details can be found in the schematics, that finally got published after they were sitting around on them for a few months.

Not publishing the schematics and a couple of other things are actually violating the grand LINARO 96boards compliance spec, but nobody there seems to really give a f...art.

I researched the 410c schematic as suggested by tbr. I can measure the input voltage at J1 and measure the input current as a voltage across the .1Ohm resistor r77 or pads 1 and 2 of j10 (both on the bottom side of the board).

    Idle, text  15.2v   115mA  1.75W
    Idle, X     15.2v   124mA  1.89W
    1CPU, X     15.2v  ~180mA  2.74W
    2CPU, X     15.2v  ~240mA  3.65W
    4CPU, X     15.2v  ~260mA  3.95W

All of my CPU loading scenarios had much more variable current readings, I tried to record the highest value I saw on a multimeter with a ~.5s update time over 10+ seconds.

Thank you! That does put an upper ceiling on power draw. (When I measure, I either use a bench PSU or splice a DMM in line.)