专题：未竟之约：张小珺访谈录 近日在《未竟之约》栏目中，智谱AI CEO张鹏在与张小珺对话中谈到"DeepSeek梁文锋选择彻底开 源"时表示，他想得很明白。第一是他大概率不想去做to B，做服务的这些事情，他也不缺钱，也没想 到要靠这个事情挣太多钱，这是很关键的一点。 第二，他也有很执着的技术上的理想，也不想因为商业化去分担精力，集中精力集中资源来做技术探索 就好了。"我猜，当然我没有那个荣幸去跟他聊一下，是不是这个真相"。 专题：未竟之约：张小珺访谈录 近日在《未竟之约》栏目中，智谱AI CEO张鹏在与张小珺对话中谈到"DeepSeek梁文锋选择彻底开 源"时表示，他想得很明白。第一是他大概率不想去做to B，做服务的这些事情，他也不缺钱，也没想 到要靠这个事情挣太多钱，这是很关键的一点。 第二，他也有很执着的技术上的理想，也不想因为商业化去分担精力，集中精力集中资源来做技术探索 就好了。"我猜，当然我没有那个荣幸去跟他聊一下，是不是这个真相"。 新浪声明：所有会议实录均为现场速记整理，未经演讲者审阅，新浪网登载此文出于传递更多信息之目 的，并不意味着赞同其观点或证实其描述。 责任编辑：李思阳 新浪声明： ...

公众号记得加星标⭐️，第一时间看推送不会错过。 据外泌体报道，人工智能的蓬勃发展将彻底改变芯片的制造方式。几十年来，全球大部分半导体都依 赖于两家巨头的技术。随着企业寻求替代方案，包括一种名为 RISC-V 的前景广阔的开源技术，这种 情况正在发生改变。RISC-V 的发音为"risk five"，目前仍属于小众产品，但有望在未来一年迎来爆 发式增长。 目前芯片制造商主要使用的两种"指令集架构"分别是英特尔控制的x86和英国Arm公司拥有的与之竞 争的架构。它们本质上是一种基础语言，决定了处理器如何执行软件命令，进而运行从操作系统、应 用程序到大型语言模型底层算法的一切程序。 随着半导体技术日趋复杂和普及，向芯片设计人员收取使用这些专有指令的费用已成为一项利润丰厚 的业务。例如，几乎所有智能手机芯片都基于Arm架构；正是这些标准使得开发者能够轻松地为不同 的设备和生态系统构建同一应用的多个版本。该公司的指令集也越来越多地应用于数据中心中央处理 器。得益于强劲的需求，根据伦敦证券交易所集团（LSEG）分析师的平均预期，该公司截至2028年 3月的财年总收入（主要来自许可费和特许权使用费）有望达到70亿美元，高于 ...

Core Insights - Moore Threads announced the full open-source release of its self-developed 3DGS foundational library, LiteGS [1] Performance Improvements - LiteGS achieves up to 10.8 times training acceleration while maintaining equivalent quality to current optimal solutions [1] - The parameter count is reduced by more than 50% [1] - With the same parameter count, LiteGS surpasses mainstream solutions by 0.2 to 0.4 dB in PSNR metrics, with training time reduced by 3.8 to 7 times [1] Lightweight Model Efficiency - For lightweight models, LiteGS requires only about 10% of the original 3DGS training time and 20% of the parameter count to achieve equivalent quality [1]

盘前市场动向 1. 12月16日(周二)美股盘前，美股三大股指期货齐跌。截至发稿，道指期货跌0.03%，标普500指数期货跌0.13%，纳指期货跌0.23%。 | 2. 截至发稿，德国DAX指数跌0.43%，英国富时100指数跌0.45%，法国CAC40指数跌0.11%，欧洲斯托克50指数跌0.28%。 | | --- | | 3. 截至发稿，WTI原油跌1.69%，报55.71美元/桶。布伦特原油跌1.63%，报59.57美元/桶。 | 英伟达(NVDA.US)收购SchedMD，以开源战略加固AI生态护城河。SchedMD的核心业务是提供大规模计算任务调度软件，这类任务往往需要占据数据中心 服务器的巨量资源。该公司的开源技术Slurm允许开发者与企业免费使用其基础软件，公司主要营收来源于工程实施与维护支持服务。本次交易财务条款尚 未披露，英伟达承诺将继续以开源形式分发SchedMD的软件产品。英伟达表示："支持最新英伟达硬件的Slurm系统已成为生成式AI关键基础设施的重要组成 部分，被基础模型开发者与AI构建者广泛用于管理模型训练和推理需求。"这家芯片巨头正通过加大开源技术投入、强化AI生态布局，以应对 ...

SchedMD提供的软件可帮助调度大型计算任务，这类任务往往会占用数据中心服务器容量的很大一部 分。该公司的Slurm开源技术允许开发者和企业免费使用，通过提供工程与维护支持服务盈利。 英伟达 凤凰网科技讯北京时间12月16日，据路透社报道，英伟达周一表示，已收购AI软件公司SchedMD。随 着竞争的加剧，英伟达正加倍押注开源技术，并加大对AI生态系统的投资以应对挑战。 虽然英伟达以开发高速芯片著称，但它也提供一系列自有AI模型，从物理模拟到自动驾驶汽车，并将 其作为开源软件供研究人员和公司使用。英伟达专有CUDA软件是大多数开发者采用的行业标准，也是 英伟达芯片的一大卖点，使得软件成为其维持在AI行业主导地位的关键。 这笔交易的财务条款未予披露。英伟达表示，将继续以开源方式分发SchedMD的软件。 在宣布这个收购消息并发布新开源（300109）AI模型后，英伟达股价上涨1.35%。(作者/箫雨) ...

近日，2025工业操作系统大会在我市吴中区举行。工业操作系统是驱动制造系统自主运行的"核心 大脑"，是工业发展的"数字底座"。本次大会以"AI赋能、智控领航、新质跃升"为主题，聚焦工业操作 系统在技术突破、生态协同与行业落地等关键维度，深入探讨"十五五"时期我国工业操作系统产业高质 量发展的可实施路径。 国家工业操作系统检验检测公共服务平台、工业操作系统（数字制造）综合创新平台同步启动。 大会会聚院士专家、产业链领军企业代表作主题发言，共同探讨工业操作系统在"十五五"关键窗口 期的跨越发展之路，为推动新型工业化"向上向新"、保障国家产业链供应链安全稳定建言献策。现场还 举办了专题论坛、供需对接会、新品发布会等多场活动。 现场，开放原子紫金专区运营总部揭牌。该运营总部由江苏软件产业人才发展基金会、苏州市工业 和信息化局与吴中区人民政府三方共建，将联动省内软件名园、重点企业、高校院所，围绕RISC-V、 工业软件、AI模型等关键领域，形成一批具有全国影响力的开源技术成果。 ...

Summary of Key Points from Conference Call Industry Overview - The conference call discusses the domestic database industry in China, particularly focusing on the trends and projections for 2025 and beyond, highlighting the impact of government funding and market dynamics [1][4][11]. Core Insights and Arguments - **Funding and Project Timing**: In 2025, funding for domestic innovation (信创) has been reduced, leading to a concentration of project tenders in the fourth quarter, with hardware equipment gaining a larger share compared to software, especially databases. A significant surge in domestic projects is expected in 2026 [1][4]. - **Market Share Leaders**: The top three players in the domestic database market for 2025 are projected to be Huawei, Tencent, and Kingbase. Dameng has struggled due to technical planning issues, limiting its participation in early tests and competition [1][6][7]. - **Open Source Database Usage**: Open-source databases like MySQL and PostgreSQL are widely used in application system replacements, although they have compatibility issues with Oracle. Kingbase, based on the PG kernel, is favored for its minimal adaptation requirements [1][8]. - **Market Growth**: The database market size is estimated to exceed 400 billion yuan in 2025, with a year-on-year growth of 20%. The growth rate is expected to accelerate to 30%-40% in 2026 [3][16]. - **Oracle's Market Position**: Oracle remains dominant in core business systems, but its market share may be gradually eroded by domestic databases over the next two to three years due to performance concerns with domestic hardware [3][23]. Additional Important Insights - **Impact of Economic Conditions**: The overall IT investment in 2025 has been significantly reduced due to economic conditions, affecting project scales and timelines, particularly in state-owned enterprises [11]. - **Database Replacement Trends**: The replacement of databases varies across industries, with the energy and healthcare sectors lagging behind. The financial sector has the most public tender projects [17][22]. - **Challenges and Opportunities**: Domestic databases face challenges in stability and performance compared to Oracle, but there are opportunities for growth driven by national policies and digital transformation needs [22]. - **Talent Retention Issues**: Domestic database firms are struggling with talent retention as larger companies attract skilled personnel. Strategies to combat this include internal training and expanding partnerships [24]. Conclusion - The domestic database industry in China is poised for significant growth, particularly in 2026, driven by government initiatives and market demand. However, challenges related to performance, talent retention, and competition with established players like Oracle remain critical factors to address for sustained success [1][22].

Group 1 - The core theme of the "Direct to Wuzhen" Global Internet Competition is "Discovering Future New Forces and Building a New Digital Ecosystem," featuring six major tracks including AI, intelligent transportation, smart healthcare, intelligent manufacturing, smart terminals, and open-source projects [2][5] - The competition attracted 1,082 projects from 29 countries and regions, with 60 awards distributed, including 6 special awards, 6 first prizes, 18 second prizes, and 30 third prizes [2][5] - The newly established "Open Source Challenge" attracted over 600 developers, highlighting the global collaboration and innovation in the digital space [2][8] Group 2 - The competition has evolved into a comprehensive platform for technology display, idea exchange, resource connection, and industry integration, allowing global innovators to better understand China's digital economy [3][4] - The competition has successfully broken down geographical barriers, facilitating the exchange of technological wisdom, industrial needs, and talent resources across countries [4][10] - Since its inception in 2019, the competition has seen participation from over 6,500 projects across 47 countries, with nearly 40% of awarded projects securing further financing [5][10] Group 3 - Major industry players such as FAW, Xiaomi, and Haier participated, showcasing innovations like the "ZhiYuan LingQu OS" and advanced surgical robots [6][7] - International projects like ORamaVR and Brill Power demonstrated significant advancements in medical training and battery management, respectively, enhancing the competition's global profile [7][8] - The open-source project track has significantly increased participation and engagement, contributing to the development of the open-source ecosystem and fostering young talent [8][9] Group 4 - The "Digital Wings" global ecological cooperation community has established a regular operational mechanism, facilitating resource connection and industry trend discussions [9][10] - The competition has transformed from a domestic-focused event to a global platform, providing a model for open cooperation in digital ecosystem construction [4][10] - The ongoing commitment to expanding global innovation cooperation networks aims to deepen the integration of digital technology with the real economy [10]

Core Insights - Open-source technology is reshaping the financial landscape, providing companies with low-cost or free access to advanced analytical and simulation tools, particularly in the post-trade environment [1] - The release of version 14 of the Open Risk Engine (ORE) enhances analytical precision and expands tool coverage, addressing the growing demand for flexible, transparent, and high-performance risk tools [1][2] Group 1: Version Enhancements - The core of version 14 is an upgrade to QuantLib v1.40, which improves performance and consistency, ensuring ORE meets the evolving needs of global financial institutions [2] - Over 100 issues have been resolved since the last release, enhancing platform stability and accuracy across all use cases [2] Group 2: Expanded Modeling Capabilities - Version 14 extends ORE's analytical capabilities to new product classes and market areas, including support for American options on commodity futures and modeling for callable bonds and their derivatives [3] - Enhancements for bond futures include the introduction of the "Cheapest-to-Deliver" feature and total return swaps for bond futures, ensuring alignment with market practices [3] Group 3: Calibration and Analysis Improvements - New features in version 14 optimize calibration and enhance modeling consistency, including a global yield curve bootstrapping function that improves the accuracy of complex yield curve construction [4] - Additional enhancements include Delta-Gamma adjustment calibration for swaptions and improved regression techniques for modeling OIS AMC risk exposure [4] Group 4: Correlation Analysis Framework - The correlation analysis framework now allows users to generate correlations based on historical scenarios, which can be integrated into XVA analysis for a more dynamic and data-driven approach to risk exposure and valuation adjustments [5] - Improved error reporting features simplify debugging processes by automatically attributing missing fixing ID errors to transaction IDs, enhancing transparency [5] Group 5: Community-Driven Development - Since its inception in 2016, ORE has evolved through continuous feedback from a global user community, reflecting direct collaboration with practitioners, academia, and developers [6] - The updates in version 14 not only bring technical improvements but also enhance the usability of risk and pricing modeling, ensuring high-quality risk analysis is accessible to all institutions [6][7]

Core Insights - Tenstorrent, led by Jim Keller, has achieved a valuation of $3 billion in its recent funding round, focusing on scalable AI accelerators for cloud and edge computing [2] - The company is developing RISC-V CPUs and plans to license its designs to other entities, utilizing open-source technologies to reduce costs associated with high-bandwidth memory [2][3] - The first chip, Tensix processor, is produced in collaboration with GlobalFoundries and features a grid-based architecture for efficient tensor computation [4] Product Development - The Grayskull processor, Tenstorrent's first Tensix processor, simplifies programming and enhances scalability, featuring up to 120 Tensix cores and supporting various AI precision formats [3] - The Wormhole variant of Grayskull reduces the number of cores to 80 but increases SRAM capacity and adds support for more precision formats, with a local memory capacity of 12GB GDDR6 [3][4] - Tenstorrent's RISC-V CPU, Ascalon, is set to compete with Arm's Neoverse V2 and V3, with performance metrics of 10 to 20 SPECint2006/GHz [5] Market Strategy - Tenstorrent is focusing on IP licensing, with existing clients including LG and Hyundai, and aims to allow customization of its cores, which is uncommon in the industry [4][5] - The company plans to release new core generations every 18 months, with the next generation, Babylon, expected to enhance performance and efficiency [7] - Tenstorrent is exploring markets in industrial and medical sectors, which have traditionally been overlooked by competitors like Arm [7] Technological Advancements - The upcoming Tensix-Neo core adopts a cluster architecture for improved area efficiency and utilization, allowing for more modular and flexible designs [8][10] - Customers will have the option to customize certain parameters while Tenstorrent manages more complex configurations, offering innovative licensing for those interested in deeper customization [10]