Radxa Cubie A7Z is the little brother of the Cubie A7A SBC, still based on a powerful Allwinner A733 octa-core Arm Cortex-A76/A55 SoC, but offered in a more compact form factor inspired by the Raspberry Pi Zero.
The compact single board computer also comes with up to 16GB RAM, a microSD card slot, optional UFS flash, micro HDMI and USB-C DisplayPort video output, a WiFi 6 and Bluetooth 5.x wireless module, a 4-lane MIPI CSI camera connector, a PCIe Gen3 FPC connector (at first for Pi Zero-sized board), and a 40-pin GPIO header.
Radxa Cubie A7Z specifications:
- SoC – Allwinner A733 (A733MX‑HN3)
- CPU
- Dual-core Arm Cortex-A76 @ up to 2.00 GHz
- Hexa-core Arm Cortex-A55 @ up to 1.79 GHz
- Single-core RISC-V E902 real-time core
- GPU – Imagination Technologies BXM-4-64 MC1 GPU with support for OpenGL ES 3.2, Vulkan 1.3, OpenCL 3.0
- VPU
- 8Kp24 H.265/VP9/AVS2 decoding
- 4Kp30 H.265/H.264 encoding
- AI accelerator – 3 TOPS NPU
- CPU
- System Memory – 1GB, 2GB, 4GB, 8GB, 12GB, or 16GB LPDDR4/4x
- Storage
- MicroSD card slot
- Footprint for optional 64GB, 128GB, 256GB, or 512GB 2-lane UFS 3.0 module
- Video Output
- Micro HDMI 2.0b port up to 4Kp60 resolution
- USB-C port with DisplayPort Alt mode
- Camera interface – 4-lane MIPI CSI connector
- Networking – Dual-band WiFi 6 and Bluetooth 5.4 module (Quectel FCU760K) + IPEX antenna connector
- USB
- USB 2.0 OTG Type-C port (power and data)
- USB 3.1 OTG Type-C port with DisplayPort Alt. mode
- Expansion
- 40-pin color-coded GPIO header (soldered or not, depending on SKU)
- PCIe Gen3 x1 FPC connector
- Misc
- USB BOOT button
- Fan header
- Power Supply
- 5V/1A via USB-C connector or
- 5V via GPIO pins 2 & 4.
- Dimensions – 65 x 30 mm
So they’ve managed to integrate most features of the credit card-sized Radxa Cubie A7A SBC into a much smaller PCB. Not bad… The Allwinner A733 SBC is probably the most powerful Raspberry Pi Zero-sized board designed so far.
Software support includes Debian Linux, Android 13, and a hardware access/control library for both operating systems. The Cubie A7Z is not listed on the documentation website just yet, but it will be about the same as for the A7A. In the meantime, you can find the build script and Debian image on GitHub. They also highlight AI support with the documentation for the Vivante NPU and Cubie ACUITY SDK already available online.

Radxa says the 1GB RAM version will sell for $15, and the Cubie A7Z is listed on AliExpress and Arace, but out of stock in both. The company guarantees the availability of the Radxa Cubie A7Z until at least September 2034. More details may be found on the product page.

Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress. We also use affiliate links in articles to earn commissions if you make a purchase after clicking on those links.






Shipping will be 15$.
The known trick to say our board is cheap, but shipping is expensive.
For me on ali the board is slightly more expensive and the shipping is less. Anyway the shipping cost usually doesn’t change with the model so it’s not exactly “a trick”. And in any case you’re supposed to compare prices as price+shipping+taxes since depending on brands, vendors, countries of origin and shipping companies, all of this can vary a lot.
You are right, but If the final price of this board is ~30$ i’m guessing there are other more cost effective SBCs.
Oh yes definitely! Depending on what you’re looking for, even the lichee-rv nano can be great and are super cheap for example (probably the cheapest functional linux boards to date).
No, I meant for 35$, I can buy RPI 4 1 gb. From Sparkfun.
This board is way more powerful than a RPi4. Also, add shipping, RPi boards aren’t delivered for free.
I wouldn’t aim for a 1 GB SKU anyway, especially with a SoC this powerful. 2 GB is realistically the lowest you’d want to cover a wide variety of use cases.
It’s sold for ¥119 CNY in China with shipping included.
In US on arace site it js $51 for 8gb model and $120 for shipping lol. Cause of tariffs I’m sure. Won’t be buying any more of these.
From where I can buy
For those very rare people that live outside the US that is not the case.
Quectel FCU760K is AICSEMI AIC8800D40 for anyone wondering.
It supports WPA3-SAE, but not 256-bit ciphers.
Thank you!
“Radxa told CNX Software that a Debian 11 with Linux 5.15 will be released in the next few hours/days, and support for Linux 6.6 and mainline U-boot is being worked on.”
Posted on August 19, 2025
@TomCubie @Jean-Luc Aufranc (CNXSoft) where we can see all this work from Radxa?
https://github.com/radxa/kernel/commits/allwinner-aiot-linux-5.15/
Images you’ll find somewhere below github.com/radxa-build and in half a decade when/if linux-sunxi community did it’s usual awesome work then these devices work somehow with mainline kernel (as usual).
Unfortunately what the sunxi community work really does is giving customers the impression that Alllooser hardware is well supported and worth buying, when that’s clearly not the case.
Not because sunxi is doing a bad job (they are really talented devs), but because neither Alllooser nor Radxa support those efforts apart from sending a couple of dev samples perhaps, which cost them nothing compared to the value they’re getting in return. Yet somehow Tom believes this is the way to “change the status-quo”.
Why should Allwinner even care about SBC (seriously bad computers) or open source enthusiasts/evangelists? They serve the lower end of the ‘Android e-waste’ market, hack together drivers for the IP blocks they’re using and forward port their BSP code to more recent kernel versions if necessary (their recent BSP is based at 6.6 at the moment, 5.15 is the former one).
If you want superiour mainline support from Allwinner those people who buy Allwinner devices (these ‘as cheap as possible’ thingies) won’t buy them anymore anyway since twice the price (software development costs money). In other words: Allwinner’s business model wouldn’t work any more and they would left their markets.
As for Radxa I think they do their best at the software front (though wrt communication they epically suck), transforming smelly BSPs into something that can at least be put on a public git repo. If you want them to do proper mainlining work you’re telling them to go out of business since ‘software development costs money’ and their boards would be twice as expensive.
And the average SBC customer is clearly not a Linux enthusiast but just cheap. If Windows on ARM would run flawlessly on all those cheap SBC I bet not much Linux users being amongst their buyers any longer. SBC (for consumers) is all about cheap and that goes not well with superiour software support. Simple as that.
RPi probably being the exception since ‘economies of scale’…
Then what’s the point in releasing general-purpose SBCs with Android e-waste hardware? To stain the brand’s name even more?
Better open-source and mainline involvement from Allwinner was part of Radxa’s marketing for the new Cubie series.
> If you want them to do proper mainlining work you’re telling them to go out of business since ‘software development costs money’ and their boards would be twice as expensive.
Yeah it’s obviously not Radxa’s job to upstream the SoC. But:
Their boards are already more expensive that competitors like the Orange Pi for no good reason. See ROCK 5B vs OPi 5 Plus for example. The latter ended up being way more popular because it’s cheaper and has more I/O. Both have terrible vendor support, but they’re equally well supported by Armbian.
Honestly I don’t understand Radxa’s decision to throw out SBC based on tons of different hardware families but I also don’t care that much or at all. If a product doesn’t fit my needs I don’t buy it. And marketing is marketing, expecting it to be ‘true’ is a little dumb IMO.
> they’re equally well supported by Armbian.
Equally bad (they still don’t care about SMP/IRQ affinity for RK3588):
https://github.com/armbian/build/blob/main/packages/bsp/common/usr/lib/armbian/armbian-hardware-optimization
> And marketing is marketing, expecting it to be ‘true’ is a little dumb IMO.
You’d be surprised how many people are still buying into the marketing claims for their other boards. And when you try to open their eyes, they go into denial: “But it’s new hardware, it takes time!” Five years old new.
Of course, marketing should always be taken with a grain of salt. But when it crosses into outright lies, sorry, gotta name and shame.
> Equally bad (they still don’t care about SMP/IRQ affinity for RK3588)
Isn’t this something that should be set by the end users? Depending on whether they need higher performance or higher efficiency?
How should end users know/care about such stuff when even those ‘famous’ Armbian ‘developers’ don’t?
Also performance vs. efficiency isn’t that easy since at least a decade the ‘race to idle’ concept has to be considered especially since we’re not talking about the CPU’s own consumption as main contributor in many situations (any more).
A big.LITTLE system condemned to run most tasks on the ‘efficient’ little cores all the time due to stupid settings is likely less power efficient than one that can switch to the performant ones on demand / when needed.
For example copying something to an SBC NAS at e.g. 30 MB/s (ODROID XU4 example from eight years ago with default settings) instead of 120 MB/s (exact same ODROID XU4 with appropriate settings) takes four times longer and will for sure consume more energy compared to a system sitting idle 3/4 of the same time span.
Overall, I agree that it’s Radxa’s responsibility to ensure upstream support for our SBCs. At the moment, our main effort is to push and educate Allwinner to allocate more resources in this direction. The encouraging part is that Allwinner’s management has already made open source a key part of their long-term strategy.
The reality, however, is that Allwinner has very limited Linux engineering capacity, to give you an idea, the entire company has only around 45 employees over the age of 40, including management. This is why they’ve decided to outsource most of the upstream work. From what I know, a bootloader upstream contract is close to being finalized, and by the end of this year we should have a full upstream boot chain for the T527/A733. In parallel, Allwinner is also exploring kernel upstreaming contracts and has already been in touch with Collabora and Baylibre.
Meanwhile, Radxa’s own team has been working on site at Allwinner’s HQ for the past two weeks, fixing bugs in the current 5.15 BSP kernel. As a former Allwinner employee, I still have a personal connection to the company. The positive side is that Allwinner’s management does listen to my input, and I try to bring in community feedback, including many of the comments here, when I advise them.
Allwinner doesn’t care for “Open Source”, but Radxa cares to (ab)use it for marketing, which feels like borderline scam.
My issue is all the effort put by lunix-sunxi, and now Radxa is always “working on”, but worked out nothing so far.
Radxa has the resource / capacity and does good upstreaming / mainlining (for Rockchip). Tiny part, will make a difference for sunxi.
If we get 1$ every time Allwinner “embraces” Open Source, we could sponsor the best support.
Looking forward to Jan 2026 to see what Radxa & Allwinner have done for 1 year of “Open Source”-ing.
U-boot & Linux patches of one Radxa developer:
https://patchwork-proxy.ozlabs.org/project/uboot/list/?submitter=84852&state=*&archive=both¶m=2&page=1
https://patchwork.kernel.org/project/linux-arm-msm/list/?submitter=216642&archive=both&state=*
Allwinner’s strength lies in offering the best processors in terms of price/quality ratio. Involving third-party companies will increase the cost of the final product. In this case, the popularity of Allwinner devices will decline.
I think the best strategy for the company would be to increase the openness of documentation and source code for consumer products so that the community can use them, as well as to help promote companies such as Radxa.
Core B2B products (like T536 T133 etc) can be outsourced to increase the appeal for big business.
i can’t believe what im about to say but i think even orange Pi is more responsable than radxa these days when getting a new product out
What kind of response do you need? I am here to answer all the questions about Radxa, Allwinner and Cubie series.
Does Radxa have the documentation & source code for the DRAM Controller of its Cubie series?
Why Radxa does zero upstream/mainline patches for Allwinner, but does hundreds for Rockchip and Qualcomm (USA)?
at last, a fair price for a fair device. I can’t wait to test it for emulation 😄
Thank you, what about a heatsink (with fan)?
It seems that in this tiny footprint the sbc going to heat quite a lot