Core Viewpoint - The article highlights the launch of the "Huagang" GPU architecture by Moer Thread, showcasing its advancements in AI training and high-performance graphics rendering, as well as the company's commitment to developing a comprehensive computing ecosystem based on its MUSA architecture [1][2][11]. Group 1: Product Launches and Innovations - Moer Thread introduced the "Huagang" architecture, which features a new instruction set that enhances computing density by 50% and improves efficiency by 10 times under the same manufacturing conditions [2]. - The "Huashan" chip, designed for AI training and inference, offers floating-point computing power, memory bandwidth, and high-speed interconnect capabilities that are competitive with NVIDIA's Blackwell and Hopper architectures [4][5]. - 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 [5]. Group 2: Infrastructure and Ecosystem Development - The "Kua'e" supercomputing cluster, based on the "Huagang" architecture, boasts a floating-point computing capability of 10 Exa-Flops, with a 60% utilization rate for Dense models and 40% for MoE models [5][8]. - Moer Thread upgraded its MUSA architecture to version 5.0, achieving over 98% efficiency in core computing libraries and planning to open-source key components [8]. - The company is also developing a MUSA ecosystem center to support research and innovation in cutting-edge fields such as quantum technology and AI for 6G [10]. Group 3: Market Position and Financial Performance - Moer Thread, founded in October 2020, has seen significant market interest, with its stock price increasing by 481% since its debut on the STAR Market at 114.28 CNY per share, reaching approximately 664.10 CNY per share and a market capitalization of about 312.15 billion CNY [11].
摩尔线程，发布新一代GPU架构