Core Insights - Moore Threads launched the "Huagang" architecture, emphasizing the development of a domestic AI computing ecosystem at the first MUSA Developer Conference [1][4] - The new architecture features a 50% increase in computing density and a 10-fold improvement in efficiency under the same manufacturing conditions [2] - The company aims to address challenges in chip manufacturing by introducing innovative instruction sets and asynchronous programming models [2] Product Launches - The "Huashan" chip is designed for AI training and inference, providing computational support for intelligent computing clusters, with performance metrics between NVIDIA's Blackwell and Hopper architectures [4] - The "Lushan" chip focuses on high-performance graphics rendering, achieving a 64-fold increase in AI computing performance and a 50% improvement in ray tracing performance [4] - The "Kua'e" computing cluster boasts a floating-point computing capability of 10 Exa-Flops, with a 60% utilization rate for Dense models and 40% for MoE models [4] Software and Ecosystem Development - The MUSA architecture has been upgraded to version 5.0, enhancing core computing libraries and communication efficiency [7] - The company plans to open-source key components of its computing acceleration libraries and system management frameworks [7] - New technologies include the MTX intermediate language for cross-generation GPU compatibility and a quantum computing framework [7][8] Market Context - The domestic AI chip industry is experiencing growth due to international restrictions on high-end chip imports, increasing market opportunities for local manufacturers [9] - Moore Threads, founded in October 2020, has seen its stock price rise by 481% since its debut on the STAR Market, with a current market capitalization of approximately 312.15 billion yuan [9]
摩尔线程发布“花港”架构，力推国产万卡智算生态