Cambricon(SH:688256) 21世纪经济报道·2025-09-02 15:26Core Viewpoint - The article discusses the significance of the "UE8M0 FP8" format in the context of the domestic chip industry and its potential to drive advancements in AI and computing efficiency in China [1][3][7]. Group 1: Understanding "UE8M0 FP8" - "FP" stands for "floating point," a fundamental unit in binary computing, with "FP8" representing 8-bit floating point, while "FP16" and "FP32" represent 16-bit and 32-bit respectively [3]. - "UE8M0" is a format that focuses all its skills on a specific range, achieving a data range similar to FP32 while being faster than FP8 formats like E5M2 [5]. - The "UE8M0 FP8" format sacrifices some precision for global stability, making it suitable for AI models that rely on large-scale computations [5][7]. Group 2: Suitability for Domestic Chips - The adoption of "UE8M0 FP8" is driven by the current state of domestic chip technology, which has room for improvement in manufacturing processes, speed, and power consumption [7]. - This format simplifies calculations to basic operations, significantly reducing computational load, which is crucial for domestic chips that face challenges in precision and efficiency [7]. - "UE8M0" is seen as a more open format that could help rebuild the ecosystem for domestic chips, which have historically been constrained by the dominance of NVIDIA's CUDA ecosystem [7][9]. Group 3: Market Implications and Uncertainties - The release of DeepSeek V3.1 has led to significant stock price increases for companies like Cambricon, which saw an 110% rise in August, indicating strong market interest in FP8-compatible chips [11]. - There is uncertainty regarding which companies will adopt the "UE8M0 FP8" format, as many firms have not clearly indicated their alignment with this new standard [11]. - Despite challenges, experts believe that domestic chip companies will ultimately succeed if they can demonstrate competitive performance advantages [12].