Ambient Scientific has introduced the GPX10 Pro, a cutting-edge SoC utilizing AI-native silicon technology to enhance AI inference in battery-powered edge devices. With its advanced AI engine capable of handling CNNs, RNNs, LSTMs, and GRUs locally, the GPX10 Pro delivers up to 100x improvements in power efficiency and performance compared to traditional 32-bit MCUs. Built on proprietary DigAn architecture, it optimizes matrix operations and eliminates typical processor overheads, allowing for accelerated edge AI functions like voice recognition and intelligent sensing with minimal power consumption.
The SoC hosts two sets of five MX8 AI cores, achieving up to 2,560 MAC operations per cycle and a peak AI throughput of 512 GOPs. Enhanced with 2MB of on-chip SRAM and an Arm Cortex-M4F CPU, it supports diverse sensor interfacing capabilities. The Nebula AI toolchain streamlines AI model training, with compatibility for TensorFlow and Keras, while the SenseMesh technology facilitates rapid sensor fusion. Samples of the GPX10 Pro are now available.
Source link
