Core Viewpoint - The collaboration between Moore Threads and Wuyishijie aims to create a fully domestic physical AI simulation system, leveraging Moore Threads' MTT S5000 GPU and Wuyishijie's SimOne 4.0 platform to provide autonomous and controllable core computing power and simulation solutions for the intelligent driving industry [2][3][4]. Group 1: Collaboration Details - The partnership focuses on adapting and optimizing the MTT S5000 GPU for the SimOne 4.0 platform, successfully integrating key processes such as large model perception mining and 4DGS simulation inference [4][5]. - The MTT S5000 GPU is designed for high-performance computing, featuring a single-card AI computing power of up to 1000 TFLOPS, 80GB of memory, and a memory bandwidth of 1.6TB/s, enabling it to handle complex simulation tasks [5][6]. Group 2: Performance Metrics - The combination of MTT S5000 and SimOne 4.0 demonstrates superior performance in large model perception mining, outperforming international competitors by up to 14% in response speed during the "pre-fill" phase [7]. - In 4DGS scene reconstruction, a high-quality reconstruction of a real Clip scene can be completed in just 2.5 hours, achieving real-time closed-loop simulation with 11 camera feeds and 1 LiDAR [7]. Group 3: Future Outlook - The collaboration signifies the acceleration of a deep collaborative ecosystem for domestic hardware and software, aiming to create a scalable physical AI evolution platform that can also be applied to embodied intelligent robots [9]. - Both companies plan to continue deepening their technological integration and enhance algorithm performance on the domestic computing foundation, contributing to a prosperous innovation ecosystem in intelligent driving [9].
摩尔线程“牵手”五一视界！共建全栈国产化的物理AI仿真体系