Are the RK3568 Parallel RGB Interface pins brought out to the heard? No, I don’t care if they aren’t RPi compatible, just if all the necessary pins are available.
New ROCK 3A brought to you by Radxa and Rockchip
I believe the comparisons are mostly looking at GPU performance.
Yeah that is a bit silly as the RK3568 is a single cut down G52 as think its intention is as a co-processor for AI than as much as graphics workload which it will not be great for.
I have done a bit with tensorflow and you can force it to thread and you seem to get very little gain for threading (seems x2 a little then no more) and maybe why a relatively high MP1 is provided (as mp1 is extremely unusual).
The video decoder/encoder is quite good however as it is strongly positioned for NVR.
Still the most unique element of that board is its native sata, QSGMII and pcie 3.0 x2 and really its not a general purpose SoC but a gateway soc to accompany RK3566 edge devices.
I have said my bit that the 3x native sata on a board of that cost has no no parallel that allows you to add a nic or further sata board as a upgrade but most importantly at that price starts with 3x native.
GPU wise there is no consideration as I don’t even think of that as a GPU its just a low ToPs co-processor for AI.
Someone else pointed out as I think they misunderstood my post as a promotion of the gpu as really you can ignore its a G52 even the Mesa guys call it a G52L ‘Little’
https://www.mail-archive.com/mesa-commit@lists.freedesktop.org/msg116463.html
The amazingly unique features of 3x native sata or the 1x QSGMII seems to be ignored which for me is time to call for Batman as the Joker is loose 
But hey, just opinion.
@Acidtech dunno why it replied to spikerguy but hey just a generalisation of a bit of confusion to the layout chosen.
This looks very promising! When there will be an option to preorder it?
Great news on pluggable M.2 card, I’m that one looking for such options, can You please tell more about it. Should google coral double TPU module work here? It requires two lanes on E slot.
I could never really work out what was meant with the dual card
Because this module uses two PCIe x1 connections, it is not compatible with all M.2 E-key card slots. The dual Edge TPUs also result in special power requirements that you must carefully review.
5.2 Power delivery and management
Caution: If you do not carefully consider the power demands of the ML models running on each Edge TPU, along with
the ability of your host to handle the corresponding current transients, the peak currents might cause brownouts or other abnormal behavior in the upstream power regulator.
As described in section 3.2 Power consumption, the current drawn by each Edge TPU is highly variable and depends on the model being executed. Although the average current drawn by a single Edge TPU might seem low (less than 500 mA),
it can repeatedly and rapidly spike up to 3 A, depending on the model you’re running. Of course, if you’re running both Edge TPUs simultaneously, you might see even larger combined current spikes. These spikes also occur suddenly: even a simple model can generate current transients in excess of 1 A/μs, which can last several tens of microseconds.
Ideally, your host system and M.2 socket can be designed to tolerate these higher currents, and your power supply can provide fast transient response performance. Alternatively, you may use some software strategies to mitigate the effects of the peak currents, such as the following:
● Schedule inferencing between the Edge TPUs so they do not draw peak currents simultaneously. In our testing, as little as a millisecond delay between inferences on each Edge TPU is enough to avoid excessive power rail current.
● Underclock one or both of the Edge TPUs to reduce the maximum of all current transients.
I have a Mini PCIe one running in an Intel Nuc and did read about the dual one which seems a really curious device especially as they do have problems with coolong so the dual has a rake of considerations and prob no-one is going to have a clue until tried even if you could provide that level of power.
Coral TPU gives +4 or +8 TOPs for AI as pluggable module. This makes such devices like Rock 3A competitive with jetsons in much better price. Power and cooling is not that difficult, still You have variety of options. I’m using 15A amp chargers and plug there as much as possible and stable working.
Still looking for information how much pcie lanes are assigned to N slot. If there is only one then there is no way You can run double TPU there (only one will work). Technically it’s possible to use nvme slot (there are adapters from M to N and that slot is described as 2x lane, there is also m 1x coral TPU) but I think it’s not that reasonable and You loose great storage option.
I’m very excited about this SBC, I’ll order them as they become available and plug them to my cluster with single or double TPU if that have change to work
BTW: if slot m.2 E slot is only 1x lane, still there is an option to convert it to double by placing there adapter with pci-e bridge, like this one:
This just adds some cost to solution, when it MAY work without such adapter.
I really dunno, even if the pmic can supply that wattage just haven’t a clue but guess its going to take someone to test.
ROCK3 wiki is wip, we should fix them this week while shipping the first batch 100pcs.
I have plenty of coral devices, even with one core tpu this seems to be great choice for AI cluster. I will order two R 3A when it will be available and make my own tests. When can I expect its release? I briefly checked stores from radxa site, they don’t have it yet.
Thanks! I just made order, I could get only 1 but this should be ok for tests. Shipping should take about 20 days, I will report here my results
@jack,could you see the TRM of Rock3A?it’s still from rk3399.
Is there gonna be an specific heatsink for Rock3A?
I’ve found the heatsink on the site, forgot to put it in my bundle 
Make sure the heatsink has fan and compatible with the fan header on 3A.
Hello, would it be possible to ask you for a copy or two for batocera?
We would be happy to work on a port of batocera to Rock 3A.
You can contact me directly here or by discord.
Batocera Link : https://batocera.org/
Batocera Discord : https://discordapp.com/invite/ndyUKA5
My Discord ID : liberodark#0928
Best regards
Hi! Is Manjaro support for Rock 3 / CM3 something to be expected sometime in the future? I can’t tell from this discussion if you got a sample upon your request. Thanks!
It is already there. Just now a ready image as rk356x is under heavy developement on mainline and we have images quartz64 which also working on rock3a with rock3a dtb which i have patched.
Try it. Not everyone is working on it so far.
- Hdmi out
- Hdmi audio.
- Usb 2
- Uart
- Thermal sensor.
- emmc
These are working while hdmi is working on older patch but still waiting for a cleaned stable support until we start making images for it.
Already recieved and added its dts to our kernel.