天数智芯公布四代通用GPU架构路线 国产算力商业化放量

Core Insights - TianShu ZhiXin (天数智芯) has released its fourth-generation general-purpose GPU architecture roadmap, focusing on "high efficiency, predictability, and sustainability" as the core design goals for "high-quality computing power" [2][5] - The company has introduced the "Tongyang" series of edge computing products, showcasing deep application cases across multiple industries and achievements in building an open ecosystem [2][5] Group 1: Product Development and Performance - Over the next three years, TianShu ZhiXin plans to launch multiple products based on the newly released fourth-generation architecture, continuously enhancing computing performance [5] - The TianShu TianShu architecture has achieved a computing power utilization rate exceeding 90%, which is a 60% improvement over the industry average [5] Group 2: Market Position and Applications - The TianGai general-purpose GPU is currently operating over 400 models, with the company's products and solutions serving more than 300 clients and completing over 1,000 deployments across sectors such as internet, finance, healthcare, and research [6] - The "Tongyang" series has been implemented in areas like embodied intelligence, industrial automation, business intelligence, and traffic intelligence, demonstrating the maturity and reliability of domestic computing power [6] Group 3: Ecosystem Development - TianShu ZhiXin has signed strategic cooperation agreements with various ecosystem partners, including hardware manufacturers and solution providers, to enhance the domestic AI computing ecosystem [6] - The company aims to create an open and win-win industrial ecosystem, enabling various industries to utilize AI with higher performance and convenience, thus accelerating the arrival of an era of accessible computing power [6] Group 4: Vision and Leadership - According to Liu Yunjie, an academician of the Chinese Academy of Engineering, AI computing power must balance quantity and quality, requiring breakthroughs in single-point performance and efficient collaboration between software and hardware [6] - CEO Gai Lujian emphasized the need for a fully self-developed stack to solidify the ecological foundation and define a new development paradigm through open cooperation, aiming to benefit various industries with domestic general-purpose GPU achievements [6]