Sixfab Raspberry Pi 5 AI HAT+ Brings Up to 25 TOPS AI Acceleration

Sixfab AI HAT+ for Raspberry Pi 5

Sixfab has announced two Raspberry Pi 5 expansion products built around the DEEPX NPUs: the AI HAT+ and the Edge AI Expansion Board. Both platforms aim to accelerate computer vision workloads on Raspberry Pi 5 systems locally but are designed for different deployment scenarios. The AI HAT+ is a PCIe HAT+ that is based on the DEEPX DX-M1 AI accelerator, and has the accelerator soldered directly onto the board. It plugs into the Pi 5 using the PCIe FFC cable and consumes power from the 40-pin header. The entry-level model Sixfab AI HAT+ comes with a DEEPX DX-M1ML accelerator rated at 13 TOPS with 512MB LPDDR4X memory, and the top model is equipped with a DX-M1M accelerator configured at 25 TOPS with 1GB LPDDR4X memory.

The Edge AI Expansion Board is designed as an under-board expansion platform while the AI HAT+ is not. It supports a hot swappable DEEPX DX-M1 M.2 accelerator module offering up to 25 TOPS at INT8 precision, and dedicated M.2 slots for NVMe storage and LTE/5G connectivity. The board incorporates a triple-M.2 design with distinct slot configurations for AI acceleration, storage, and cellular communications, enabling all three subsystems to run concurrently. AI HAT+ is for prototyping and development, while Edge AI Expansion Board provides AI acceleration with NVMe storage and LTE/5G network access for real-world deployments.

Previously, we covered several Raspberry Pi 5 HATs, including the SupTronics X1208, Waveshare PCIe HAT+, and Waveshare PCIe to USB 3.2 HAT+. Feel free to check them out if you’re interested.

Sixfab AI HAT+ for Raspberry Pi 5 Front Side

Sixfab AI HAT+ Specifications:

• Compatible SBC: Designed for the Raspberry Pi 5 platform.
• AI Acceleration Options:
  • DEEPX DX-M1M NPU delivering up to 25 TOPS (INT8) with 1GB LPDDR4X dedicated memory.
  • DEEPX DX-M1ML NPU delivering up to 13 TOPS (INT8) with 512MB LPDDR4X dedicated memory.
• Host Connectivity: PCIe Gen3 x1 interface connected through a 16-pin FFC cable.
• Cooling and Hardware Features:
  • Passive cooling solution included by default.
  • Onboard 2-pin JST connector for optional fan installation.
• Power Requirements:
  • Powered directly from the Raspberry Pi 5’s 40-pin GPIO header.
  • Input rating: 5V / 3A.
  • No external power connector required.
  • A 27W Raspberry Pi power supply is recommended, as the standard 15W PSU does not provide sufficient power.
• Power Consumption:
  • NPU Peak Load: 2.5–3W during intensive AI inference.
  • NPU Idle: Approximately 0.5–1W.
  • Total Raspberry Pi 5 + AI HAT+ Consumption: Around 13–15W under typical workloads.
• Physical Characteristics:
  • Dimensions: 65 × 56.5 mm
  • Height: 6.56 mm
  • Fully compliant with the Raspberry Pi HAT+ form factor specification.
• Operating Temperature: 0°C to 70°C (commercial-grade range).
• Compliance and Certifications: Certification status includes CE, FCC, UKCA, RoHS, and REACH compliance (currently in progress).

Sixfab AI HAT+ RPi5

The Edge AI Expansion Board is compatible with the Raspberry Pi OS (Trixie). It uses the HAT+ EEPROM for auto-configuration and setting it up is as simple as installing the dxrt-runtime package from Sixfab’s APT repository, which includes the driver and runtime. Both products (the AI HAT+ and the Edge AI Expansion Board) run the DEEPX software stack and support deployment of models compiled using the DXNN SDK. For more details, refer to the documentation and the community page.  

The Sixfab AI HAT+ for Raspberry Pi 5 with DX-M1 AI accelerator is now available on the Sixfab store: 13 TOPS (DX-M1ML) variant for $63 and 25 TOPS (DX-M1M) variant for $90. The Edge AI Expansion Board is listed as coming soon, and no pricing information is available at this time. 

Images used courtesy of Sixfab.