Sixfab AI HAT+ for Raspberry Pi 5 integrates DEEPX DX-M1 AI accelerator
Sixfab’s new AI HAT+ brings DEEPX DX-M1 acceleration to Raspberry Pi 5, offering 13 or 25 TOPS for computer vision at $63 and $90. It sips power, hooks up via PCIe, and runs YOLOv8, MobileNet, and ResNet locally—but no LLMs yet.
Condensed by AI-Portable from Editorial queue.
Vision AI at a Low Price
The Sixfab AI HAT+ for Raspberry Pi 5 integrates DEEPX’s DX-M1 NPU directly onto a PCIe HAT+ board, connecting via the Pi’s FFC ribbon and drawing power from the 40‑pin header. It comes in two flavors: the DX-M1ML with 13 TOPS (INT8) and 512 MB LPDDR4X memory, and the DX-M1M with 25 TOPS and 1 GB. Under full inference, the NPU sips just 2.5–3 W, leading to a combined Pi 5 + HAT+ consumption of 13–15 W—a 27 W PSU is required because the 15 W unit can’t feed both. The board measures only 65 × 56.5 mm and sits 6.56 mm tall, with a passive heatsink and a fan header for extra cooling. Setup is automatic via the HAT+ EEPROM once you install the dxrt-runtime package from Sixfab’s APT repository on Raspberry Pi OS (Trixie).
Software, Models, and Limitations
Sixfab supplies a Model Zoo with pre‑compiled networks for immediate use:
- YOLOv8 for object detection
- MobileNet for classification
- ResNet for recognition
Custom models can be exported to ONNX, then compiled to DXNN format with the DX-COM tool. The runtime supports both Python and C++ APIs, so existing pipelines migrate easily.
However, the accelerator is strictly a vision‑AI engine. It lacks transformer decoder support and has at most 1 GB of NPU memory, so it cannot run LLMs or generative tasks. The Hailo‑8‑based Raspberry Pi AI HAT+ shares this limitation, but the newer AI HAT+ 2 (Hailo-10H) delivers up to 40 TOPS and 8 GB memory for LLM/VLM applications. Sixfab says LLM capability is on DEEPX’s silicon roadmap, but no timetable exists.
Pricing and Roadmap
Pricing is competitive: $63 for the 13‑TOPS variant, $90 for the 25‑TOPS version. Both are available now from the Sixfab store. The company is also building an Edge AI Expansion Board for the Pi 5 that will marry the DX-M1 accelerator with NVMe storage and LTE/5G cellular connectivity—an all‑in‑one edge AI appliance that could suit remote camera deployments. Details are forthcoming.
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.
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