Embedded electronics specialist Sipeed has introduced the approaching launch of a high-performance “AI-Field” for on-device machine studying workloads: the MaixBox-M4N, which pairs the Maix-IV M4N system-on-module with the M4N-Dock provider board..
“M4N-Dock is a device-side mixed-precision high-computing edge computing field launched by Sipeed,” the corporate writes of its newest design, delivered to our consideration by Linux Gizmos. “It’s geared up with [Axera’s] third-generation high-energy-efficiency clever imaginative and prescient chip AX650N, with built-in AI [Artificial Intelligence] computing energy of 43.2 TOPS [tera-operations per second] @ INT4 or 10.8TOPS @ INT8 [precision].”
Sipeed’s newest launch is the Maix-IV M4N module, which it claims can hit over 40 TOPS for edge AI workloads. (📷: Sipeed)
The system-on-module itself is, as Sipeed says, powered by the Axera AX650N, which incorporates eight Arm Cortex-A55 cores working at as much as 1.7GHz, a neural processing unit (NPU) coprocessor with assist for INT4, INT8, INT16, FP16, and FP32 precision and delivering a claimed 43.2 TOPS of compute on the lowest INT4 precision, and a picture sign processor (ISP) with 8k30 assist. There’s 8GB of LPDDR4x, which the person can cut up between the CPU and NPU cores, and 32GB of eMMC storage.
For exterior connectivity the M4N “Core Board” system-on-module is designed to be related to the M4N-Dock provider, which incorporates two HDMI 2.0 outputs with 4k60 assist, two SATA 3.0 ports and one M.2 SATA slot, two four-lane MIPI Digital camera Serial Interfaces (CSIs), one PCI Categorical Gen. 2 lane on a mini-PCIe slot, one USB 3.2 Gen. 1 and three USB 2.0 ports, RS485 and RS232 serial ports, and two gigabit Ethernet ports — although, considerably unusually, no easily-accessible general-purpose enter/output (GPIO) pins.
The provider board comes bundled with a heatsink and fan meeting to tame the highly effective system-on-chip. (📷: Sipeed)
“[The] NPU has highly effective efficiency,” Sipeed claims of the system’s capabilities. “When utilizing just one/3 of the computing energy, that’s, 3.6 TOPS, the commonly-used open supply AI fashions carry out as follows: YOLOv5s can attain 130 frames [per second] (7.66ms) below 640×640 decision picture enter. MobileNetv2 reaches 1798 frames [per second] (0.556ms) below 224×224 decision picture enter.”
The corporate has opened orders on its AliExpress retailer for the module at $155 plus transport, with a module-and-dock bundle priced at $230 together with a heatsink and fan meeting for cooling; the MaixBox-M4N bundle, in the meantime, features a rugged aluminum case at $248. Extra data is obtainable on the Sipeed wiki.