上市15天，摩尔线程剑指英伟达腹地

Core Viewpoint - The article discusses the recent developments by Moore Threads, including the launch of their new GPU architecture "Huagang" and AI training and inference chip "Hushan," highlighting their ambition to build a comprehensive ecosystem similar to NVIDIA's CUDA [2][4]. Group 1: New Technologies and Innovations - Moore Threads introduced the intermediate language MTX, part of the MUSA 5.0 software stack, which allows compatibility across different generations of GPUs, significantly reducing adaptation costs for developers [5][7]. - The "Huagang" architecture represents a substantial upgrade, with a 50% increase in computing power density compared to the previous "Pinghu" generation, and it supports full precision calculations from FP4 to FP64 [11][10]. - The new architecture also features asynchronous programming and supports large-scale interconnections, enabling the expansion of intelligent computing clusters to over 100,000 nodes [11][10]. Group 2: Chip Launches and Performance - Two new chips were launched based on the "Huagang" architecture: "Lushan," designed for AI training and inference, and "Hushan," aimed at high-performance graphics rendering [14][13]. - The "Lushan" chip can adapt to various protocols and supports expansion with up to 1,024 super nodes, while the "Hushan" chip boasts a 64-fold increase in AI computing performance and a 50-fold improvement in ray tracing capabilities [16][14]. Group 3: Ecosystem Expansion - The company is focusing on building a self-sufficient ecosystem while exploring new boundaries, exemplified by the introduction of the "Changjiang" SoC, which combines CPU, GPU, and NPU to provide up to 50 TOPS of heterogeneous AI computing power [21][20]. - Moore Threads is also developing the MT Robot intelligent solution and the MT Lambda simulation training platform, aiming to integrate various engines to enhance research and development efficiency [23][25]. - The company's strategy reflects an ambition to establish a competitive ecosystem in the GPU industry, akin to NVIDIA's, while facing significant challenges in building a robust domestic ecosystem [26][25].