国泰海通表示，据DeepSeek官微，DeepSeek-V3.1 使用了 UE8M0 FP8 Scale 的参数精度， UE8M0 FP8是针对即将发布的下一代国产芯片设计，国产芯片有望得到大规模应用。新一代的国产芯片纷纷开 始支持FP8，包括沐曦690、芯原VIP9000、寒武纪690、摩尔线程 MUSA 3.1 GPU、海光深算三号等。 我们认为，随着国产AI芯片设计技术、制造工艺等方面的提升，且随着国产大模型的持续发展以及对 国产芯片的适配和支持，国产算力的市占率有望持续提升，国产芯片的行情有望持续推进。 每日经济新闻 （责任编辑：董萍萍 ） 【免责声明】本文仅代表作者本人观点，与和讯网无关。和讯网站对文中陈述、观点判断保持中立，不对所包含内容 的准确性、可靠性或完整性提供任何明示或暗示的保证。请读者仅作参考，并请自行承担全部责任。邮箱： news_center@staff.hexun.com 创业板人工智能ETF国泰（159388）跟踪的是创业板人工智能指数（970070），涨跌幅为20%，该 指数从创业板市场中选取涉及软件开发、智能硬件制造、大数据处理等人工智能技术及相关应用的上市 公司证券作为指数 ...

Core Viewpoint - The introduction of the UE8M0 FP8 format and the next-generation domestic chips signifies a major advancement in the Chinese AI industry, indicating a shift towards hardware-software synergy and reduced reliance on foreign computing power [1][36]. Group 1: Market Reaction - Domestic chip companies' stock prices surged, with Cambrian Technology rising nearly 14% and hitting the daily limit, becoming the top company on the STAR Market [2][35]. - The semiconductor ETF also experienced a significant increase of 5.89% within half a day [5]. Group 2: Technical Explanation of UE8M0 FP8 - UE8M0 FP8 is based on the MXFP8 format, which was defined in the Open Compute Project's 2023 specification, allowing for an 8-bit micro-scaling block format [10][11]. - The MXFP8 format allows for block-level scaling, which retains the 8-bit width while expanding the dynamic range significantly [14][22]. - The UE8M0 format specifically uses an all-exponent representation for scaling factors, which simplifies processing and reduces computational overhead [22][23]. Group 3: Implications for Domestic Chips - Most existing domestic AI accelerators still use FP16/BF16 + INT8 pathways, but new chips like MUSA 3.1 GPU and VIP9000 NPU are beginning to support native FP8 [26]. - The UE8M0 format can save 75% of bandwidth compared to traditional FP32 scaling, making it a crucial optimization for next-generation architectures [27]. - The support for UE8M0 FP8 by various domestic chip manufacturers indicates a significant competitive advantage and potential for improved performance in AI applications [36][37]. Group 4: Industry Collaboration - DeepSeek has collaborated with 15 manufacturers to validate the UE8M0 format, with several companies already adapting their products to support this new standard [26][29]. - The collaboration between DeepSeek and domestic chip manufacturers is likened to the historical Wintel alliance, suggesting a strong ecosystem is being built around these technologies [37].

Core Viewpoint - The release of DeepSeek's V3.1 model highlights a significant shift in the domestic AI ecosystem, emphasizing the collaboration between software and hardware through the introduction of the UE8M0 FP8 floating-point format designed for next-generation domestic chips [6][11]. Group 1: Technical Innovations - DeepSeek's V3.1 model features a mixed reasoning architecture, improved thinking efficiency, and enhanced agent capabilities [6]. - The UE8M0 FP8 format prioritizes numerical range stability over decimal precision, allowing for stable training on domestic GPUs, which differ from NVIDIA's hardware [11][13]. Group 2: Competitive Landscape - The competition around FP8 standards represents a broader struggle for computational ecosystem dominance, with NVIDIA solidifying its position through the Blackwell architecture and MXFP8 format [13]. - Domestic AI firms are encouraged to innovate collaboratively from models to chips, as seen with companies like Muxi and Suiyuan Technology optimizing FP8 computation efficiency [13][18]. Group 3: Future Trends - The collaboration between hardware and software, even at the cost of temporary performance compromises, is likely to become a prevailing trend in the industry [14]. - The UE8M0 FP8 format signifies a critical step for the domestic AI industry, moving from isolated breakthroughs to comprehensive stack collaboration, which may offer more long-term value than merely increasing model scale [16].

Group 1 - The core viewpoint is that China's computing power infrastructure is rapidly advancing, positioning the country as a global leader in digital infrastructure by mid-2025, with a total computing power scale ranking second worldwide [1] - As of June 2023, China has established 4.55 million 5G base stations and has 226 million gigabit broadband users, indicating significant growth in digital infrastructure [1] - The development of artificial intelligence is heavily reliant on computing power, which is considered a fundamental "energy" for AI advancements [3] Group 2 - Computing power is categorized into supercomputing, general computing, and intelligent computing, with intelligent computing being crucial for the iterative development of AI technologies [4] - China's total computing power is projected to reach 280 EFLOPS by the end of 2024, with intelligent computing expected to grow over 40% by 2025, reaching 90 EFLOPS, which constitutes 32% of the total [5] Group 3 - Recent advancements in hardware and software capabilities have been highlighted, including Huawei's "Ascend 384 Super Node" and the introduction of competitive domestic GPU products, indicating a robust ecosystem for computing power [6] - The integration of the integrated circuit industry has progressed, creating a complete supply chain from design to manufacturing, enhancing the usability of domestic computing power [6] Group 4 - The energy consumption of data centers is a significant concern, with predictions indicating that by 2030, China's data center energy consumption will exceed 400 billion kilowatt-hours [7] - Various regions are implementing energy-efficient practices, such as using liquid cooling technology and renewable energy sources, to enhance the energy efficiency of computing power centers [8] Group 5 - By the end of 2024, over 9 million standard racks will be in use across data centers in China, with an average Power Usage Effectiveness (PUE) of 1.46, showcasing improvements in energy efficiency [8] - The overall outlook for China's computing power infrastructure is positive, with rapid market growth, expanding demand scenarios, and accelerated technological advancements, which are essential for establishing a competitive advantage in the AI era [8]

Core Viewpoint - The release of DeepSeek V3.1 and the mention of a new architecture and next-generation domestic chips have caused significant excitement in the AI industry, leading to a surge in stock prices for domestic chip companies like Cambricon, which saw an intraday increase of nearly 14% and became the top company on the STAR Market [4][22]. Group 1: UE8M0 FP8 Concept - The term "UE8M0 FP8" can be broken down into two parts, with "UE8M0" representing a scaling factor in the MXFP8 path, which is defined in the Open Compute Project's specification for 8-bit micro-scaling formats [7][8]. - MXFP8 is based on FP8, compressing conventional floating-point formats to 8 bits, allowing for a significant expansion of the dynamic range while maintaining an 8-bit width [8][15]. - The scaling factor in UE8M0 consists of 8 bits, which can be allocated to sign, exponent, and mantissa bits, with the "U" indicating unsigned [11][12]. Group 2: Benefits of UE8M0 FP8 - UE8M0 allows processors to restore data using simple operations, significantly reducing the complexity of floating-point multiplication and normalization, thus shortening critical clock paths [15][17]. - The dynamic range of UE8M0 spans from 2^(-127) to 2^(128), providing ample space for subsequent block scaling and reducing information loss while maintaining 8-bit tensor precision [15][17]. - The adoption of UE8M0 can lead to a 75% reduction in data traffic compared to traditional FP32 scaling, making it a crucial optimization direction for next-generation architectures [18][27]. Group 3: Domestic Chip Manufacturers - Several domestic chip manufacturers, including Cambricon, Hygon, and Moore Threads, are preparing to support FP8, with Cambricon's chips already being compatible with FP8 calculations [22][23]. - The market has reacted positively to the potential of these domestic chips, with the STAR 50 index rising by 3%, marking a three-and-a-half-year high for the chip industry [24][27]. - The collaboration between DeepSeek and domestic chip manufacturers represents a shift towards a more self-sufficient AI ecosystem in China, reducing reliance on foreign computing power [27][28].

Core Viewpoint - The article highlights the significant performance improvement of HanGuangJi, a leading AI chip company in China, with its stock price nearing that of Kweichow Moutai, indicating strong market interest and growth potential in the AI chip sector [2]. Group 1: Company Performance - HanGuangJi's stock reached 1243.20 yuan, with a trading volume exceeding 15.8 billion yuan, reflecting robust market activity [2]. - The company reported a revenue of 9.89 billion yuan and a net profit of 2.81 billion yuan in Q4 of the previous year, marking its first quarterly profit since going public [2]. - In Q1 of this year, HanGuangJi's revenue surged to 11.11 billion yuan, a significant increase from 0.26 billion yuan in the same period last year, with a net profit of 3.55 billion yuan [2]. Group 2: Industry Developments - DeepSeek recently launched DeepSeek-V3.1, which is designed for the upcoming generation of domestic chips, marking a step towards the agent era in AI [2]. - Several AI chip manufacturers, including Moole Technology and Suiyuan Technology, have introduced chips that support FP8 precision computing, enhancing performance in large model training by 20% to 30% [3][4]. - The domestic AI industry is increasingly utilizing local computing power, with companies like iFLYTEK testing large models based on fully domestic computing resources [4]. Group 3: Market Outlook - Research reports from various brokerages indicate a stable growth in the semiconductor supply chain, with rising wafer foundry capacity and a positive outlook for the semiconductor market [5]. - The introduction of new precision formats by DeepSeek suggests a growing application of domestic AI chips in training and inference processes, especially in light of international supply chain fluctuations [5].

Core Viewpoint - DeepSeek has officially open-sourced its next-generation model DeepSeek-V3.1 on the Hugging Face platform, marking a significant step towards the era of intelligent agents, with notable enhancements in tool usage and task capabilities through Post-Training optimization [1] Technical Upgrades - The new model features a context window increased from 64K to 128K, enabling it to process long texts equivalent to 300,000 Chinese characters, which supports long document analysis, complex code generation, and deep multi-turn dialogues, resulting in a performance improvement of approximately 40% in tool calling and complex reasoning tasks [2] - The architecture has transitioned from a single reasoning mode to a dual-mode architecture, enhancing support for complex task processing and multi-step reasoning, with the introduction of DeepSeek-Chat for quick responses and DeepSeek-Reasoner for logical reasoning and problem-solving [3] - Enhanced tool calling capabilities allow for more reliable interactions with enterprise systems, introducing a strict mode to ensure output format compliance, thus facilitating smoother integration with internal APIs and databases [4] Chip Compatibility and Market Impact - DeepSeek-V3.1 utilizes a parameter precision format called UE8M0 FP8, designed for upcoming domestic chips, which significantly reduces memory usage and computational resource demands compared to traditional formats [6][7] - Domestic AI chip manufacturers such as Cambricon, Huawei Ascend, and others are expected to benefit significantly from this optimization, with noticeable stock price increases following the announcement [8] Competitive Landscape - The open-source model poses challenges to international closed-source model providers like OpenAI and Anthropic, as DeepSeek-V3.1's performance and cost advantages may compel these companies to adjust their API pricing or disclose more technical details [11] - The open-source strategy of DeepSeek, which allows free commercial use and modification under the Apache 2.0 license, contrasts sharply with the limited open-source approach of competitors, fostering a more competitive environment and enabling smaller companies to access advanced model technologies at lower costs [13][14] Beneficiaries of Open Source - Companies developing applications based on large models, cloud computing and hardware vendors, and traditional enterprises with data and application scenarios are expected to benefit from the open-source model, leading to increased demand for GPU computing power and facilitating digital transformation [14] - The rise of open-source models will create a more complex competitive landscape, with other open-source model providers needing to keep pace with the performance benchmarks set by DeepSeek-V3.1 [15] Developer Ecosystem - The open-source model encourages global developer participation, allowing for personalized customization and optimization of the model, which can lead to rapid performance improvements [19] - Companies must weigh the benefits of open-source versus closed-source models, with open-source providing cost savings and greater autonomy, particularly for small to medium-sized enterprises focused on technology independence [20]

Group 1 - The Sci-Tech Innovation Index (000680) has seen a strong increase of 4.54%, with constituent stocks such as Haiguang Information (688041) and Yuntian Lifeng (688343) both rising by 20.00% [1] - The Taikang SSE Sci-Tech Innovation Board Composite Index Enhanced A (023970) achieved a weekly return of 5.92%, ranking first among enhanced products of the Sci-Tech Innovation Index [1] - The DeepSeek model has been updated to version 3.1, now supporting FP8 precision and domestic chips, which is expected to accelerate the domestic computing power ecosystem [1] Group 2 - Companies like Huawei and Cambrian are expanding their market competitiveness in the domestic chip sector, with Huawei's Ascend chips being increasingly applied in government and financial sectors [1] - Other domestic AI chip manufacturers are also showing accelerated penetration, with Muxi providing N-series GPUs focused on cloud AI inference scenarios [1] - There is a recommendation to pay attention to the related industrial chain benefiting from the domestic AI chip localization [1]

Core Viewpoint - The release of DeepSeek V3.1 and its mention of the next-generation domestic chip architecture has caused significant excitement in the AI industry, leading to a surge in stock prices of domestic chip companies like Cambricon, which saw an intraday increase of nearly 14% [4][29]. Group 1: DeepSeek V3.1 and UE8M0 FP8 - DeepSeek V3.1 utilizes the UE8M0 FP8 parameter precision, which is designed for the upcoming generation of domestic chips [35][38]. - UE8M0 FP8 is based on the MXFP8 format, which allows for a more efficient representation of floating-point numbers, enhancing performance while reducing bandwidth requirements [8][10][20]. - The MXFP8 format, defined by the Open Compute Project, allows for a significant increase in dynamic range while maintaining an 8-bit width, making it suitable for AI applications [8][11][20]. Group 2: Market Reaction and Implications - Following the announcement, the semiconductor ETF rose by 5.89%, indicating strong market interest in domestic chip stocks [4]. - Cambricon's market capitalization surged to over 494 billion yuan, making it the top stock on the STAR Market, reflecting investor optimism about the company's capabilities in supporting FP8 calculations [29][30]. - The adoption of UE8M0 FP8 by domestic chips is seen as a move towards reducing reliance on foreign computing power, enhancing the competitiveness of domestic AI solutions [33][34]. Group 3: Domestic Chip Manufacturers - Several domestic chip manufacturers, including Cambricon, Hygon, and Moore Threads, are expected to benefit from the integration of UE8M0 FP8, as their products are already aligned with this technology [30][32]. - The anticipated release of new chips that support native FP8 calculations, such as those from Huawei, is expected to further strengthen the domestic AI ecosystem [30][33]. - The collaboration between DeepSeek and various domestic chip manufacturers is likened to the historical "Wintel alliance," suggesting a potential for creating a robust ecosystem around domestic AI technologies [34].

Group 1 - Multiple stocks in the computing power sector experienced significant gains, with SMIC rising 6.29% and Chipone increasing by 5.39% [1] - AI-related stocks saw even larger increases, with Haiguang Information up 17.19%, Zhongke Shuguang up 10%, and Cambricon up 12.4%, reaching a new high of over 1170 CNY per share and a market capitalization exceeding 490 billion CNY [1][3] Group 2 - Cambricon, a leading domestic AI chip company, reported improved performance, achieving a revenue of 989 million CNY and a net profit of 281 million CNY in Q4 of last year, marking its first quarterly profit since going public [3] - In Q1 of this year, Cambricon's revenue surged to 1.111 billion CNY, a significant increase from 26 million CNY in the same period last year, with a net profit of 355 million CNY [3] Group 3 - DeepSeek recently launched DeepSeek-V3.1, which is designed for the upcoming generation of domestic chips, utilizing UE8M0 FP8 precision [3][4] - Several domestic AI chip manufacturers are supporting FP8 precision calculations, with companies like Moore Threads and Suiyuan Technology introducing chips that enhance performance in large model training by 20% to 30% [4] Group 4 - Recent reports from various brokerages highlight the stability and growth of the semiconductor supply chain, with rising wafer foundry capacity and a continued favorable semiconductor market [5] - The potential for increased application of domestic AI chips in training and inference based on the DeepSeek model was noted, especially in light of international supply chain fluctuations [5]