Sixfab has launched the AI HAT+ for Raspberry Pi 5, a PCIe HAT+ based on the DEEPX DX-M1 AI accelerator, which we also found in the DEEPX DX-AIPlayer, Mini DX-M1 SoM, and ALPON X5.
Unlike the M.2 module used in the ALPON X5, the AI HAT+ has the accelerator soldered directly to the board. It connects to the Pi 5 via the PCIe FFC cable and draws power from the 40-pin header. The board is also available in 13 TOPS and 25 TOPS versions and is designed to run vision AI tasks such as object detection and segmentation locally on the Pi 5.
Sixfab AI HAT+ specifications:
- Supported SBC – Raspberry Pi 5
- AI Accelerator (one or the other):
- DEEPX DX-M1M with up to 25 TOPS (INT8), 1 GB LPDDR4X NPU memory
- DEEPX DX-M1ML with up to 13 TOPS (INT8), 512 MB LPDDR4X NPU memory
- Host Interface – PCIe Gen 3 x1 via 16-pin FFC cable
- Misc – Passive cooling by default; included 2-pin JST fan connector
- Power Supply
- Input – 5V / 3A via Pi 5 40-pin GPIO header (no auxiliary connector required); Note: 27W PSU required, and the 15W PSU is insufficient
- Consumption
- NPU peak – 2.5–3 W under full inference load
- NPU idle – ~0.5–1 W
- Combined Pi 5 + HAT+ – 13–15 W (27W or more PSU recommended)
- Dimensions – 65 x 56.5 mm (Raspberry Pi HAT+ compliant), 6.56 mm height
- Temperature Range – 0 to 70°C (Commercial)
- Certifications – CE, FCC, UKCA, RoHS, REACH (currently in progress)
The board works with Raspberry Pi OS (Trixie). It uses the HAT+ EEPROM for auto-configuration, and setup only requires installing the dxrt-runtime package from Sixfab’s APT repository, which includes the driver and runtime.
You can either use pre-compiled AI models from the Sixfab Model Zoo, such as YOLOv8, MobileNet, and ResNet, or run your own models. Custom models can be exported to ONNX and compiled into DXNN format using the DX-COM tool. The runtime supports both Python and C++. You’ll find more details in the documentation.
In terms of AI power, the Sixfab AI HAT+ is similar to the Raspberry Pi AI HAT+ (Hailo-8), as both are designed for object detection and image processing. But the Sixfab board is not designed for generative AI and cannot run LLMs as it lacks transformer decoder support and has limited on-board memory. The main differences between the two are in their architecture, software stack, and pricing. In comparison, the newer AI HAT+ 2 (Hailo-10H) is designed for both computer vision and generative AI workloads, with higher performance (up to 40 TOPS) and 8GB of dedicated memory to support LLM and VLM applications. SifFab says that “LLMs are on the DEEPX silicon roadmap” and the company will support them as the silicon enables, but no dates were provided.
The Sixfab AI HAT+ for Raspberry Pi 5 with DX-M1 AI accelerator is available now on the Sixfab store: the 13 TOPS (DX-M1ML) variant is priced at $63, while the 25 TOPS (DX-M1M) variant is priced at $90. The company is also working on the Edge AI Expansion Board for Raspberry Pi 5, which combines AI acceleration with NVMe SSD storage and LTE/5G cellular connectivity on a single board, but details are limited at this stage.
Debashis Das is a technical content writer and embedded engineer with over five years of experience in the industry. With expertise in Embedded C, PCB Design, and SEO optimization, he effectively blends difficult technical topics with clear communication
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.
