maix4-hat Axera's AX650N/C Octa-Core Cortex – A55 A compact AI hat from the Raspberry Pi 5 built around a Raspberry-IV core board delivering up to 72 TOPS (INT4) or 18 Tops (INT8) with CPU, SOC, and 8K video encoding/decoding support.
It features 8 GB LPDDR4X RAM and 32 GB EMMC storage, PCIE 2.0, HDMI 2.0A (4KP60), USB 3.0, USB 2.0, multiple camera inputs, SPI LCD, I²C touch, speakers, fan connectors and more. Designed for plug-and-play use on Raspberry Pi 5 and other boards, it accelerates transformer-based models and is ideal for space-constrained edge AI tasks in smart cameras, industrial inspections, and multimodal AI applications.
siped maix4-hat spec:
- soc – axera ax650n
- CPU – Octa-core arm Cortex-A55 @1.7 GHz Neon Support
- NPU – 43.2 Tops @int4, 10.8 tops @int8 supports INT4, INT8, INT16, FP16, and FP32 inputs. According to Sipeed/Axera, it is equivalent to the Nvidia 40 top.
- ISP – Up to 8kp30 (8,192 x 4,320 @ 30 fps). Maximum resolution 16,384 x 16,384
- DSP – 800 MHz Dual-Core DSP
- Video Decoding – H.264/H.264 Video Decoder 8kp60, Max 32x Channels @1080p30
- Video Encoding – H.264/H.265 Video Encoder up to 8kp30, up to 32x Channels @1080p30
- System Memory – 8GB 64-bit LPDDR4X (2GB for system + 6GB with AI)
- Storage
- 32 GB EMMC 5.1 Flash
- microSD card support
- screen
- MINI HDMI 2.0A to 4K @60 fps
- 10-pin FPC SPI display interface with 6-pin FPCI²C touch interface
- Camera – 0.8 mm 4-pin USB Camera Interface
- audio
- 2-pin speaker connector
- Onboard microphone
- USB
- USB 2.0 (480 Mbps) Type-C OTG port
- USB 3.0 (5 GBPS) Type-A port
- Debug – USB Type-C Port
- Expansion – 1 x PCIe 2.0 (1 lane, 16-pin FPC), Raspberry Pi 5 compatible
- others
- Reset, Boot 0, and Boot 1 buttons
- 1.25mm 2-pin cooling fan header
- Blue and white LED indicators on the core board
- Power supply – 5V from PI or other board
- Dimensions – 65 x 56 mm
M4N-HAT supports a complete AI development stack and works on Sipeed's updated Maixpy platform with appropriate documentation. Developers can use pre-trained AI models using Axera's Pulsar2 tool to embrace faces and transform or deploy their own models. It supports common AI tasks such as image recognition, object detection, and even running large-scale language models. The board is fully compatible with the Raspberry Pi 5, with setup and usage guides available in the AXCL documentation. Detailed information such as hardware and software documentation, SDK downloads, AI development resources (AXCL, Pulsar2), model hubs, sample code repository, and more are available on the Wiki.
The benchmark chart compares 18 top Maix4-hat with the RK3588 with 6 top NPU, hailo-8 (26 tops), and hailo-8L (13 tops) across a variety of AI models. We have long pointed out that the number of tops offered by manufacturers does not always move towards actual results. In RAW FPS performance, Maix4-hat always outperforms most CNNs and all others in trans workloads, reaching up to 5,961 FPS and 5,073 FPS on Sqeezenet11 on MobileNETV2.
The bottom chart shows performance as a percentage compared to MAIX4-HAT (which is set to 100% on all models. If another chip runs at 50%, it is only half faster than that model's Maix4-hat. In most cases, the other accelerators (RK3588, hailo-8, hailo-8l) are less than half of maix4-hat. The only big exception is the hailo-8 Resnet50. Here, the hailo module beats maix4-hat at about 131% of the speed. So, overall, MAIX4-HAT is usually much faster, especially for computer vision models, but it may be possible for another chip to be faster on certain model types.
There are plenty of other AI hats for the Raspberry Pi 5, including the official Raspberry Pi AI Hat+, Pineboards AI Bundle (Hailo 8L), and the Raspberry Pi AI Kit. All of these can perform on-device AI acceleration for vision, voice and other ML tasks.
Sipeed Maix4-hat is now available for purchase Aliexpress for $173It used to be on Sipeed's Aliexpress Store $149.
Debashis Das is a technical content writer and embedded engineer with over five years of experience in the industry. With his expertise in embedded C, PCB design and SEO optimization, he effectively blends difficult technical topics with clear communication
Supports CNX software! Donate via cryptocurrency, become a patreon sponsor, or buy items on Amazon or Aliexpress. Also, if you purchase after clicking on these links, you will use the affiliate links in the article to win the committee.
