Jetson Thor计算平台

Search documents
英伟达发布 Jetson Thor:2070TFLOPS 算力赋能机器人新“大脑”
Jing Ji Guan Cha Wang· 2025-08-26 06:59
北京时间8月25日晚间,英伟达在产品发布媒体会上官宣了此前预热的"机器人新大脑"——专为物理AI 与机器人应用设计的Jetson AGX Thor计算平台,开发者套件售价3499美元,量产模块批量采购单价降 至2999美元。这款搭载Blackwell架构GPU的"机器人新大脑",以2070TFLOPS的AI算力、128GB显存及 全栈软件支持,重新定义了边缘计算性能标准。 发布会伊始,英伟达正式宣布其专为机器人、自动驾驶计算平台设计的系统级芯片Thor量产上市。Thor 这一命名灵感来源于北欧神话中的雷神,寓意着强大的力量,而该芯片也确实不负其名,在性能上实现 了巨大飞跃。与上一代Orin芯片相比,Thor芯片采用了与英伟达AI数据中心GPU相同的新一代Blackwell 架构。 作为"机器人新大脑",Jetson Thor计算平台集成了CPU、GPU、内存、连接与电源管理模块。算力性能 方面,据环球网报道,在FP8标准(低精度的8位浮点数格式,业内常用16位浮点数格式FP16或FP8衡量 AI算力)下,其单颗AI算力达到1035TFLOPS(每秒万亿次浮点运算),在FP4标准下更是高达 2070TFLOPS ...
国泰海通|机械:富士康与英伟达联手,有望打造人形机器人具身智能落地标杆
国泰海通证券研究· 2025-06-25 15:12
Core Viewpoint - Nvidia and Foxconn are negotiating the deployment of humanoid robots at Foxconn's new Houston factory, aiming for completion by Q1 2026, marking a significant milestone in the industrialization of humanoid robots [1] Group 1: Foxconn's Manufacturing Expertise - Foxconn leverages its extensive experience in manufacturing as a leading global electronic manufacturing service provider to accelerate smart manufacturing transformation [2] - At GTC 2025, Foxconn showcased diverse applications of its robotic technology, including semiconductor composite robots that provide intelligent transportation solutions for the semiconductor industry [2] - Foxconn is deepening its technical collaboration with Nvidia to optimize production processes through diversified AI solutions, indicating a strong demand for AI technology in new production lines [2] Group 2: Nvidia's Role in Humanoid Robot Development - Nvidia is driving the development of the humanoid robot industry with its multimodal models and computing ecosystem, introducing Project GR00T, which integrates large language models and visual perception technology [3] - The Jetson Thor computing platform, designed specifically for robots, efficiently runs multimodal generative AI models, enhancing the capabilities of humanoid robots [3] - Nvidia collaborates with various companies, including Boston Dynamics and Hexagon, to create a comprehensive ecosystem from training to deployment, exemplified by Hexagon's humanoid robot AEON, which utilizes Nvidia's accelerated computing [3] Group 3: Significance of Collaboration - The collaboration between Nvidia's AI models and Foxconn's manufacturing capabilities is expected to break through application bottlenecks for humanoid robots and expand into other industrial sectors [4] - Successful implementation of this application scenario could facilitate the transition of manufacturing from automation to autonomy, creating a broader market space for autonomous manufacturing [4]