China Mobile Cloud Unveils Heterogeneous Hybrid Inference System: Brain-Inspired Chip Paired with Domestic GPU Shatters Single-Architecture Performance Barriers
21 hour ago / Read about 0 minute
Author:小编   

At the inaugural event of the Computing Power China Expo, Du Yujian, the Technical Director of the Brain-Inspired and Optical Laboratory at China Mobile Cloud, presented a cutting-edge heterogeneous hybrid inference system tailored for large-scale model inference tasks. This system leverages an innovative 'brain-inspired chip + domestic GPU' architecture, positioning itself as a formidable competitor to NVIDIA's Vera Robin + Groq solution. By integrating Lingxi Technology's brain-inspired chip with Iluvatar CoreX's GPU, the system not only achieves more than double the performance boost compared to similar GPU clusters in DeepSeek V4 Flash inference scenarios but also cuts operational costs by over 40%, ensuring self-sufficiency and enhanced controllability. The incorporation of brain-inspired computing capabilities offers robust support and complements existing GPU resources, paving the way for a new era in high-performance computing.