Arm China Introduces 'Zhouyi' X3 NPU IP, Establishing a New Standard for On-Device AI Computational Efficiency
1 day ago / Read about 0 minute
Author:小编   

Arm China has recently introduced its latest-generation NPU IP, named the 'Zhouyi' X3. This innovative product leverages a DSP+DSA architecture tailored for large-scale models, enabling seamless collaborative processing of CNN and Transformer algorithms. It is complemented by the 'Zhouyi' NPU Compass AI software platform. The 'Zhouyi' X3 provides AI computational cores for four key sectors: infrastructure, intelligent vehicles, mobile devices, and smart IoT. The hardware architecture has undergone a significant upgrade, transitioning from fixed-point to floating-point computation. It integrates a dedicated AI hardware engine, AIFF, which substantially boosts computational efficiency and density. The software ecosystem is designed to optimize the end-to-end performance of large-scale models and supports custom development tailored to client needs. Through the synergy of software and hardware, the 'Zhouyi' X3 achieves a remarkable performance enhancement, with multi-core computing linearity reaching 70%-80% and delivering outstanding performance in handling large-scale models. This versatile product finds broad applications across various AI devices, enhancing Arm China's product portfolio in the on-device AI market.