The world of single board computers (SBCs) has always been dominated by a few familiar names – Raspberry Pi being the obvious example. But every so often, a new player emerges with a design that makes the community sit up and pay attention. That’s what’s happening right now with the Allwinner A733, also known informally as the A7z.
This SoC is starting to appear in developer leaks, short videos, and early testing boards. While not yet backed by the ecosystem of a Raspberry Pi, it has a spec sheet that’s hard to ignore: a hybrid CPU design, a capable GPU, a dedicated NPU, and even a RISC-V coprocessor for real-time workloads.
For those of us in IoT, M2M, and edge computing, this chip could signal a new wave of affordable but powerful alternatives to the Pi 5 and Rockchip-based boards.
The heart of the chip: hybrid CPU design
The Allwinner A733 is built around a high-performance, hybrid octa-core design:
- 2 × Cortex-A76 performance cores, clocked up to 2.0 GHz
- 6 × Cortex-A55 efficiency cores
This is similar in spirit to what we see on modern smartphone SoCs – a big.LITTLE cluster arrangement that balances raw performance with low power consumption.
Compared to the Raspberry Pi 5, which carries 4 × Cortex-A76 cores at around 2.4 GHz, the Allwinner design trades two fewer performance cores for a more power-efficient cluster of A55s. In practice, this means:
- Single-thread performance will be strong thanks to the A76 cores.
- Light, multi-threaded workloads may run more efficiently across the A55 cluster.
- Heavy multi-threaded performance could lag behind the Pi 5’s full quad A76 setup.
It’s not just about raw cores, though. The SoC is designed to balance workloads, making it attractive for IoT applications where efficiency matters as much as throughput.
GPU: modern graphics and compute
The A733 is equipped with an Imagination BXM-4-64 MC1 GPU. On paper, this supports:
- OpenGL ES 3.2
- Vulkan 1.3
- OpenCL 3.0
That’s a more modern graphics feature set than the Broadcom VideoCore VII found in the Raspberry Pi 5. For graphics acceleration, UI rendering, or compute workloads that can be offloaded to OpenCL, the Allwinner GPU should prove more versatile.
The real question is software support. Raspberry Pi’s Broadcom stack is mature, while Allwinner historically relies heavily on community contributions to get the most out of its GPUs. For industrial or embedded use, though, Vulkan and OpenCL support could unlock interesting edge compute projects.
Built-in AI acceleration: a 3 TOPS NPU
One of the standout features is the integrated Vivante VIP9000 NPU, delivering up to 3 TOPS at INT8. It supports mixed precision across INT8, INT16, FP16, and BF16, which makes it suitable for a wide range of machine learning inference tasks.
For AI-driven IoT projects – think image recognition at the edge, voice analysis, or anomaly detection – this is a genuine differentiator. Raspberry Pi has no built-in NPU, meaning you’d normally add an external accelerator like Google Coral or Intel Myriad. The Allwinner chip builds it in.
Framework support looks promising too: ONNX, TensorFlow, PyTorch, Caffe, TVM. If this is backed by working SDKs, the NPU could be the A733’s killer feature.
RISC-V coprocessor: low-power flexibility
Another unusual addition is a RISC-V E902 coprocessor running at 200 MHz. This sits alongside the main CPU and is optimised for low-power and real-time tasks.
Why does this matter? It allows the board to handle lightweight monitoring or real-time control jobs without waking the main A76/A55 cluster. In industrial IoT, where efficiency and uptime matter, this could help reduce power draw and improve responsiveness.
How does it compare to Raspberry Pi 5?
This is the inevitable question. The Raspberry Pi 5 set a new bar for hobbyist SBCs with its quad A76 CPU and improved GPU. Here’s a side-by-side comparison:
| Feature | Allwinner A733 / A7z | Raspberry Pi 5 |
|---|---|---|
| CPU | 2 × Cortex-A76 (2.0 GHz) + 6 × Cortex-A55 | 4 × Cortex-A76 (2.4 GHz) |
| GPU | Imagination BXM-4-64 (Vulkan 1.3, OpenCL 3.0) | Broadcom VideoCore VII (OpenGL ES, Vulkan support limited) |
| NPU | 3 TOPS (INT8, FP16, BF16) | None |
| MCU | RISC-V E902 coprocessor | None |
| Software support | To be established, depends on Radxa/community | Mature Raspberry Pi OS ecosystem |
| Ecosystem | Likely Radxa Cubie A7z first | Huge Pi ecosystem |
So, in pure CPU terms, the Pi 5 probably holds the edge for multi-threaded tasks. But once you factor in the integrated NPU, modern GPU, and RISC-V flexibility, the Allwinner A733 starts to look like a very strong option – particularly for AI and edge compute projects.
Likely boards: Radxa Cubie A7z and friends
The A733 isn’t just a concept – it’s already being whispered about in relation to Radxa and Cubieboard. TikTok and YouTube videos are already mentioning a Radxa Cubie A7z board, which looks set to be one of the first commercial SBCs using this SoC.
Radxa has a history of partnering with Allwinner, and they’re known for solid hardware at competitive prices. If they bring the A7z to market with decent RAM, storage, and I/O options, it could slot neatly into the space currently occupied by Raspberry Pi alternatives like the Rockchip RK3588-based boards.
Real-world applications
Where could this board make a difference?
- AI at the edge: Local inference on video streams, sensors, or speech, without external accelerators.
- IoT gateways: Balancing efficiency and power with the hybrid CPU design.
- Media and UI projects: Benefiting from modern GPU features.
- Industrial monitoring: Using the RISC-V coprocessor for low-power real-time tasks.
- Education and development: Offering a new architecture to experiment with, especially in AI workloads.
The verdict
The Allwinner A733 / A7z is still fresh, with only early glimpses appearing in the wild. But what we’ve seen so far suggests this could be the most exciting Allwinner release in years.
For hobbyists, it may offer a cheaper way into AI workloads than adding accelerators to a Pi. For professionals in IoT and edge computing, it could be an interesting alternative platform – if the software support is up to scratch.
Raspberry Pi still wins on community and ecosystem, but this is a genuine competitor. If Radxa or Cubieboard deliver stable hardware and drivers, the Allwinner A7z could carve out a niche as the go-to SBC for AI-driven projects in 2025.