Here you can see the first prototype of H3-OLinuXino. Our first Quad core OSHW Linux SBC.
What it have:
- H3 quad core Cortex-A7
- 1GB DDR3 Ram
- optional 4GB NAND Flash
- micro SD card
- USB-OTG
- 2x USB-HOST
- LCD connector with 0.1″ step (not assembled on the pictire)
- GPIO connector with 0.1″ step (not assembled on the pictrure)
- UEXT connector
- HDMI
- Ethernet 100Mb
- dimensions only 80×50 mm
Now these first prototypes will be put on heavy testing before we run the board in production. We want to see if they will be able to run Linux yet or just Android.
From what I’ve read on the net H3 is overheating big time, due to the integration of the Ethernet PHY to the SOC silicon. We will see how bad it is.
We have no price for H3-OLinuXino yet as there are just first prototypes and there may be changes in the design.
UPDATE: 20:00 o’clock – It works! … but even with aluminum heatsink it heats up to 80C 😦 we will take more experiments and measurements but so far it seems H3 consumes x3 times more than A20
olimex 
Nov 13, 2015 @ 16:13:32
Nice! I’m looking forward to having a box that’ll play h.265.
Nov 13, 2015 @ 16:20:15
yes, it is way too hot @ 1.5 GHz. Chip easily reaches 90+ degrees Celsius and starts to underclock and shutdown cores.
See what is written here:
https://linux-sunxi.org/Orange_Pi_PC#CPU_clock_speed_limit
Nov 13, 2015 @ 16:29:03
BTW, I do not see power connector.
Board can be powered from the USB-OTG port?
It really looks very low cost design (maybe except the 1GB DDR3): no PSU, no PMU, no external Ethernet PHY.
It would be nice if those savings bring the price below 30 euro 🙂
Nov 13, 2015 @ 16:32:45
you are very humble, I was thinking that the prognosis will start from EUR 8.99 😉
Nov 13, 2015 @ 22:20:16
My guess for price is in the range of A20-LIME2 😉
Nov 13, 2015 @ 16:38:18
Without an annoying fan the H3 should be clocked with 960 MHz maximum and a quality heatsink is also needed. The best starting point to get Linux on this thing is here: http://www.orangepi.org/orangepibbsen/forum.php?mod=viewthread&tid=342&extra=page%3D1
Nov 13, 2015 @ 17:02:02
even PIC with ethernet PHY works at 60C I guess AW underestimated the impact to move the ethernet PHY on the silicon
Nov 16, 2015 @ 07:17:40
Either provide adequate cooling or set up clocking to what it can actually handle. Sure, these chinese manufacturers oversell any day. Just like everyone else does these days…
Dec 16, 2015 @ 03:37:52
…after looking on photo of Orange Pi PC I’ve got idea they are using axe to separate boards. Just take a look on zoomed view. Separate glass fibers are falling apart. Ouch!
Nov 13, 2015 @ 16:26:26
It’s very nice. Debian please.
Nov 16, 2015 @ 07:40:04
> It’s very nice. Debian please.
I would rather tell “mainline kernel support please” (as well as u-boot). Once this one ready, it’s not a major issue to debootstrap own image, etc. But before it happens, there is some trouble on the way.
Sure, one can boot Debian with vendor SDK kernel like 3.10, BUT it tends to have long-term stability issues, outdated, some chinese drivers really bugged, etc. Then, they can use some proprietary components like it happened to Cedar lib.
So at the end of day, it is moving target and imperfect world, and we have to live with it 😛
Nov 16, 2015 @ 10:00:36
Mainline work is on the way (see https://groups.google.com/forum/#!searchin/linux-sunxi/h3/linux-sunxi/XYSGx4k8nIE/NFdB2LtqBgAJ). I guess by the time this board hits the shelf there will be mainline versions for kernel and u-boot (maybe not stable). There is still the issue about the mali gpu and free libs so I don´t know about h.265, maybe with android it´s possible.
Nov 13, 2015 @ 17:22:23
Speaking of ‘Quad core OSHW Linux SBC’… any news on the potential A20 successor? But based on the experiences with the last quad core SoCs I played with (H3 and Actions Semi’s S500/ATM7059 — both suffering from overheating problems) I have mixed feelings when Allwinner puts two more cores on the die.
Nov 13, 2015 @ 17:30:51
ahhh the 4 core A20 upgrade will be ready probably second half of 2016, Allwinner actually were starting working on this when they told me and I didn’t understand correctly so blogged about it then they were not happy that I didn’t kept this in secret until they announce themself 🙂
Nov 13, 2015 @ 17:42:45
Understandably. 🙂 Back to the H3: I would strongly recommend using RPi-Monitor to get graphs for internal temperatures, cpufreq and DRAM freq and count of CPU cores (since Allwinner’s kernel code shuts CPU cores down). Data sources:
/sys/class/thermal/thermal_zone0/temp
/sys/class/thermal/thermal_zone1/temp
/sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_cur_freq
/sys/devices/platform/sunxi-ddrfreq/devfreq/sunxi-ddrfreq/cur_freq
expr 1 + `awk -F”: ” ‘/^processor/ {print $2}’ < /proc/cpuinfo | tail -n1`
It's easy to adjust the templates. I did it for A20 in the past and just recently did it for the S500 (this beast and its PMU exceed easily 100°C under full load, I even managed to trigger emergency shutdowns at 125°C): http://wiki.linux-xapple.org/w/index.php/RPi-Monitor
Nov 13, 2015 @ 20:00:33
Alas, on a “do-everything” OLinuXino board, the H3’s integrated Ethernet PHY can’t shine because the cost savings from integration are masked by the additional costs of the many other bits that are added. “Do-everything” carries a price penalty.
To make an H3-based OLinuXino do what the SoC is best at (ie. low cost Ethernet), the board should have only the H3 itself + essential circuitry, RJ45 with magnetics, RAM, a GPIO connector and UEXT, nothing else. (Perhaps 1 USB port for the plug’n’play peeps, as it’s cheap.) Then it would be the perfect extremely low-cost wired IoT board, and the price would be able to reflect the H3’s high level of integration. Price down, sales up.
Call it H3-OLinuXino-NET to emphasize that it’s entirely dedicated to Ethernet-based control applications (and not HI), and you’d have a new product for your “IoT” category which currently doesn’t have a wired member.
Morgaine.
Nov 13, 2015 @ 22:10:19
There seems to be H3-OLinuXino-NANO in the works, see next blog post
Nov 16, 2015 @ 07:28:11
Uhm, well, in IoT you usually do not need 4 CPU cores. What exactly you want to compute on 4 cores? Though it does little harm…
As for phys, its something really strange. I’ve took a look on quite a some various ICs and these vary wildly. Some crappy ARM SoC @ mere 250MHz with integrated phy is so hot I can’t touch it (CNS2132; partially supported by openwrt, but utter crap overall). Some 100M switch ICs like ADM6996 are very hot as well, even if I do not get why it have to be like this. Because… because, say, some usual Realtek switch which contains whole 5 gigabit ports at most getting a bit warm. So it seems some PHY implementations just happen to be very inefficient and it really depends on implementation.
Nov 16, 2015 @ 17:56:51
“…“IoT” category which currently doesn’t have a wired member.”
Sure it does – only it’s hiding in the “Olinuxino” category: the RT5350F-OLinuXino, more wired ports than one could possibly need, all you need are the magjacks… 😉
Nov 13, 2015 @ 20:32:49
By the way, integrating Ethernet PHY doesn’t *have* to make a chip hot. I have some TI TM4C1294-based LaunchPads which integrate 10/100 Ethernet PHY on their microcontroller SoC, and their normal operating temperature with Ethernet active isn’t much warmer than ambient. I guess it must depend on the specific implementation.
Morgaine.
Nov 13, 2015 @ 20:52:13
I guess these temp measurements are occurring when running Linux of some kind ? There might be some flaw in Linux that causes over-heat somehow, or perhaps a hardware issue where the are bus-conflicts, or similar effect ? Bus conflicts essentially cause temporary but recurring shorts for the chip. Suggest turning things off or removing components to find if there is a conflict at play, somewhere.
Although mentioning an opposition board is not ideal, does anyone know if the PCduino-8 (it has an Allwinner H-8 chip – i,e, 8 x cores), is overheating too? Its picture shows no heatsink – – –
I have read articles where is even where A-20’s are using everything at full speed (DDR, GPU & CPU) it will need a HeatSink. Perhaps Allwinner are cutting corners – – –
Something may be unseen here – – – although the idea of the onchip-PHY complicating the H3 die too much is a good explanation..
* *
Nov 14, 2015 @ 13:26:34
The ‘problem’ with the H3 is that it’s being marketed as “up to 1.6 GHz” but that’s just marketing bullsh*t suitable to provide higher Antutu scores. When you do benchmarking for your tablet’s press release you also switch off the ‘thermal controls’ contained in Allwinner’s 3.4 kernel: Normally the CPU cores are being shut down one after the other. To work with this chip in a realiable way you simply don’t exceed a core voltage of 1.2V which translates to the H3 being a 1.0 GHz chip (try 960 MHz as the maximum clock speed and you’re safe). Again: https://linux-sunxi.org/Orange_Pi_PC#CPU_clock_speed_limit
Nov 16, 2015 @ 07:45:58
You’re right: A20 could get pretty hot when running without heatsink for a while doing some intense computations. E.g. try “7z b” or “openssl speed” commands and take a look on CPU temperature. It would get kinda hot.
Furthermore, claims about 1.5GHz are crap and you better count on it to be able to do 1 GHz. Sure, you can clock it to 1.5GHz but it would overheat and lose stability or would be extremely unstable. On other hand, even 1GHz with 2 CPU cores isn’t bad for many uses. It happens to be reasonable as NAS/microserver like thing, thanks to SATA, and cheaper, faster, and overall better replacement is yet to be seen in this area (yeah, SATA beats USB 2.0 any day).
Nov 13, 2015 @ 22:11:28
Interesting !
Amazingly, as a pure coincidence, I’ve received my first OrangePI-PC yesterday.
It cheap price of $15 attracted me. I’ve tested it a bit, and it is well faster than my A20. The only caveat I’ve found : no SATA.
Nov 14, 2015 @ 18:12:58
Allwinner H3 has display LCD?
Nov 16, 2015 @ 14:48:49
Any signs of Android 5.1 for the H3?
Nov 16, 2015 @ 15:22:44
the internet to/from China is quite unreliable, from Allwinner try to upload H3 SDK to our ftp for a long time but they can’t complete, same happen with the cloud service they use, we start to download but the file is huge 7-8GB and always broke unfinished, so I have no access to latest H3 SDK to check what android it have
Nov 16, 2015 @ 15:37:14
Secret software always leads to a giant mess. If they’d put it on a public git server we could sync to the US hosted AOSP servers and then only pull in the deltas from them. Lobbing tarballs is no way to develop software and it is a security patch nightmare.
Amlogic has this nice git system, but they won’t give me access.
http://openlinux.amlogic.com/Android/Mbox
Since I have no AMLogic access I working on the RK3128 instead. Firefly has their Android 5.1 SDK up in a public git repo. Rockchip is also making considerable effort to mainline their kernel using their own employees.
Nov 16, 2015 @ 15:44:21
Have them make a new gmail account which will give them 15GB of free storage. Upload it with GDrive. Apparently that is not too hard to do from China since Google has servers in Asia. Once it is GDrive it is easy to get to from anywhere. But it is still a tarball with no way to get security patches.
Nov 16, 2015 @ 17:24:42
Gmail is not accessable in China
Nov 16, 2015 @ 17:37:32
Gmail is not ‘officially’ accessible in China. VPNs are commonly used to bypass the great firewall. Firefly is using one to put the Rockchip Android distributions onto GDrive.
Nov 16, 2015 @ 17:54:25
You should think about setting up a fex server for such situations (it has been made for exactly that kind of stuff): http://fex.rus.uni-stuttgart.de/index.html (more technical details: https://fex.rus.uni-stuttgart.de/doc/concept )
Nov 16, 2015 @ 19:48:54
If you run Linux on FTP, can access command line and able to install programs, try “axel” program. This small 40Kb program makes wonders. It splits file to blocks and downloads them in parallel, achieving remarkable speed. But best of all, it proven to be extremely robust to all kinds of network errors. I once faced weird misconfigured server, which breaks connections for something like every 40-120Kb, which is far worse than most of chinese servers could be. Yet I managed to get 4Gb file AND it even passed md5sum verification. Kinda impressive, granted HTTP does not haves any checksms on its own. and when connection breaks it is really up to downloader program to be smartass enough.
Nov 17, 2015 @ 11:02:17
If it is on baidu there are tools to help you download better.
But I honestly do not think it’s changed anything from the first / second release. probably the usual kernel 3.4.39 and Android 4.4
Nov 16, 2015 @ 19:09:57
Hi Im from TURKEY 😉
What are you think about this boards price? Can you give a little info plase.
I work with OrangePi PC. H3 is working with 75C. H3 is very hot. Your board is come with heatsink? Maybe chage the freq. andwork 1gz !
Thanks.
ARMTurkey.com
Nov 17, 2015 @ 13:26:52
I think that would be good: A20, 2GB RAM, wifi integrated and NAND flash. And reduced price.
Nov 19, 2015 @ 15:30:14
I asked in the OrangePi forums whether people can do a little benchmarking with ‘default’ settings (overclocked: 1.53GHz cpufreq and 672 MHz DRAM) and more sane settings (1008/480 MHz). When the H3 is not actively cooled with both heatsink and fan then it performs identical with both settings since overheating/throttling occurs with default/overclocked settings: sysbench time 183.7 vs. 184.6 seconds
http://www.orangepi.org/orangepibbsen/forum.php?mod=redirect&goto=findpost&ptid=724&pid=6637
Nov 27, 2015 @ 00:48:33
Today I adjusted dvfs settings for H3 and was curious whether 4 low clocked cores work more efficient than 1 fast core. The quad-core attempt makes really sense: http://www.orangepi.org/orangepibbsen/forum.php?mod=redirect&goto=findpost&ptid=743&pid=6976&fromuid=29411
Dec 13, 2015 @ 21:22:23
Plase chnge H3 with A83t or A64. We hate fans and aliminium heatseak. 😉
I love A20 & Debian with ARM Boards.
Dec 14, 2015 @ 04:44:18
Why “change”? Just make all kinds of boards and let us try ’em. We’ll see what we prefer. Btw, A20 can also get pretty hot if you run it on 1GHz for a while. Try e.g. “openssl speed” or “7z b” while looking on CPU temperature. Without heatsink it can burn your skin if you touch CPU. And A20 just would not be stable at advertised 1.5GHz. Some CPUs would start up on 1.5GHz, but if you’ll dare to put anyhow strong load, it would overheat and all sorts of weird things would happen.
Dec 14, 2015 @ 10:03:36
The H3 doesn’t need a heatsink at all when the moronic dvfs settings from Xunlong are fixed. It idles at 1.5W consumption and even the 4 CPU cores are a good thing for IoT stuff: http://forum.armbian.com/index.php/topic/504-quick-review-of-orange-pi-pc/
Really looking forward to OSHW H3 boards from Olimex!
Dec 15, 2015 @ 21:13:36
Small addendum: The H3 running with kernel 4.4.0-rc4 is able to benefit from UASP (USB Attached SCSI). It’s the fastest USB 2.0 device I ever measured: 39 MB/s write and 41.5 MB/s read: http://forum.armbian.com/index.php/topic/504-quick-review-of-orange-pi-pc/#entry3244
And the H3 has 4 independent USB ports and not just 2 or 3 like most of the other recent Allwinner SoCs 🙂
Dec 18, 2015 @ 12:34:47
Agree with @Thomas: An H3 board with GbE and all 4 USB exposed would be nice!
Dec 18, 2015 @ 17:40:06
@vickycq: http://linux-sunxi.org/Sunxi_devices_as_NAS#New_opportunities_with_mainline_kernel
Looking forward to an OliNAS in 2016 with H3 😉
Sep 17, 2016 @ 14:30:32
Would be nice if there could be a variant without the big connectors assembled, similar to the A33-board, so hat the user can solder on these connectors by themselves if they need them. These big connectors are always a pain to desolder, which is necessary when the project has height constraints for the board.