国产化

Search documents
重磅!华为发布准万亿大模型
Mei Ri Jing Ji Xin Wen· 2025-05-30 11:41
Core Insights - Huawei has launched a new model called Pangu Ultra MoE, which has a parameter scale of 718 billion, marking a significant advancement in the MoE model training field on the Ascend AI computing platform [1][3][6] - The release of Pangu Ultra MoE and the Pangu Pro MoE series demonstrates Huawei's capability in achieving a fully controllable training process for domestic computing power and models, validating the innovation capacity of China's AI infrastructure [3][6] Model Architecture and Training Innovations - The Pangu team has introduced innovative designs in model architecture and training methods to address the challenges of training ultra-large-scale and highly sparse MoE models, achieving stable training on the Ascend platform [1][4] - Key innovations include the Depth-Scaled Sandwich-Norm (DSSN) architecture and TinyInit initialization method, which have enabled long-term stable training with over 18TB of data [4] - The introduction of the EP loss load optimization method ensures better load balancing among experts and enhances their specialization capabilities [4] Performance and Efficiency Improvements - The training methods disclosed by Huawei have enabled efficient integration of large sparse MoE reinforcement learning (RL) post-training frameworks on the Ascend CloudMatrix 384 supernodes [5] - Recent upgrades have improved the pre-training system's performance, increasing the multi-factor utilization (MFU) from 30% to 41% [5] - The Pangu Pro MoE model, with 72 billion parameters and 16 billion active parameters, has demonstrated performance comparable to larger models, ranking first among domestic models under 100 billion parameters in the SuperCLUE leaderboard [5] Industry Implications - The successful training and optimization of ultra-large-scale sparse models on domestic AI platforms signify a closed-loop of "full-stack domestication" and "fully controllable processes" from hardware to software, and from research to engineering [6] - This advancement provides a strong foundation for the development of China's AI industry, reinforcing confidence in domestic AI capabilities [3][6]
中信重工: 中信重工关于部分募投项目延期的公告
Zheng Quan Zhi Xing· 2025-05-30 11:21
Core Viewpoint - The company announced a delay in the completion date of its fundraising project, specifically the "Key Equipment Production Line for Panel Boxes" project, which is now set to be completed by December 31, 2026, instead of the previously planned date of June 30, 2025 [1][4]. Fundraising Overview - The company raised a total of RMB 828,462,797.65 through the issuance of 240,134,144 shares at a price of RMB 3.45 per share, after deducting issuance costs of RMB 12,484,371.91 [1][2]. - The actual net amount raised was lower than the amount initially reported to the China Securities Regulatory Commission (CSRC) in September 2023, prompting a proportional adjustment in the investment amounts for the fundraising projects [2][3]. Project Details - The total investment amount for the projects is RMB 108,801.61 million, with the adjusted amount for the "Key Equipment Production Line for Panel Boxes" project being RMB 81,597.84 million [3][4]. - As of March 31, 2025, the company had signed contracts worth RMB 15,494.48 million for the project and had utilized RMB 8,632.51 million of the raised funds [4][5]. Delay Reasons and Project Importance - The delay is attributed to the use of advanced production technologies, the need for custom equipment, and the import of certain machinery, which necessitated a cautious approach to ensure efficient use of funds and risk mitigation [4][5]. - The project is crucial for the company as it aligns with national policies promoting the development of high-end equipment manufacturing and aims to address key technological challenges in the industry [5][6]. Impact of Delay - The delay is a prudent decision based on actual project conditions and will not affect the investment direction or the implementation of the project, ensuring no harm to the interests of the company and its shareholders [6][7]. - The company will continue to monitor market changes and optimize resource allocation to ensure the effective use of raised funds [6]. Approval Process - The board of directors approved the delay without the need for a shareholders' meeting, confirming compliance with relevant laws and regulations [6][7].
重大突破!刚刚,华为发布!
券商中国· 2025-05-30 10:43
5月30日,券商中国记者从华为获悉,华为在MoE模型训练领域再进一步,重磅推出参数规模高达7180亿的全 新模型——盘古Ultra MoE,这是一个全流程在昇腾AI计算平台上训练的准万亿MoE模型。同时,华为发布盘 古Ultra MoE模型架构和训练方法的技术报告,披露众多技术细节,充分体现了昇腾在超大规模MoE训练性能 上的跨越。 中国AI产业注入强心剂。 业内人士分析,华为盘古Ultra MoE和盘古Pro MoE系列模型的发布,证明华为不仅完成了国产算力+国产模 型的全流程自主可控的训练实践,同时在集群训练系统的性能上也实现了业界领先。这意味着国产AI基础设 施的自主创新能力得到了进一步验证,为中国人工智能产业的发展提供了一颗"定心丸"。 国产算力与国产模型重大突破 据悉,训练超大规模和极高稀疏性的MoE模型极具挑战,训练过程中的稳定性往往难以保障。针对这一难题, 华为盘古团队在模型架构和训练方法上进行了创新性设计,成功地在昇腾平台上实现了准万亿MoE模型的全流 程训练。 在模型架构上,盘古团队提出Depth-Scaled Sandwich-Norm(DSSN)稳定架构和TinyInit小初始化的方法, ...
全国劳模张玉忠:电力“大工匠” 拥抱新“智变”
Zhong Guo Neng Yuan Wang· 2025-05-30 09:48
1993年进入电厂,万余次精准操作创造零差错纪录,成功处置数十起突发险情,荣获"全国劳动模范""全国工人先锋号""吉 林能源工匠""国家电投杰出奋斗者"等30余项殊荣……这是国家电投吉电股份四平第一热电公司(以下简称"四平一热公司") 发电部技能大师兼培训师张玉忠,用32年职业生涯绘就出的电力工匠"答卷"。 32载,从普通技工成长为国家级"产业导师",张玉忠将"干一行必精一行"的誓言锻造成金。为什么要长期扎根一线?他有 自己的理解:"要不是自己经历过,就没法设身处地发现问题,更别提解决行业难题了。" 破冰攻坚——零下25摄氏度的技术突围 在零下25摄氏度的渣仓铁梯旁,睫毛上结着白色冰霜的张玉忠,正试图摘下和右手虎口的血痂冻在一起的手套。东北的冬 天零下20多度是常态,要完成室外的渣仓料位的手动测量任务,需要爬上20多米高的铁梯子,铁梯经常结冰挂霜,攀爬铁梯 时,手扶栏杆的手套常被粘在栏杆上,极易发生不安全事件。 这个隐患是张玉忠一直以来的"心病",他憋着一股劲儿,下决心必须解决这个难题。从"三防"角度出发,他扎根实验室, 带领团队成员反复研究,终于研发出一套具有微正压自清洁功能的"气旋"式无触点料位监控装置, ...
中国产业叙事:北方华创
新财富· 2025-05-30 08:03
本文约 5 0 0 0 字,推荐阅读时长 2 0 分钟,欢迎关注新财富公众号。 20世纪60年代,在"两弹一星"工程与"156项重点工程"宏大叙事之下,中国半导体制造设备的国产化征程悄然启动。 1965年,北京建中机器厂(国营700厂)的成立,标志着中国首次系统性布局半导体制造设备的研发。这座藏身于北京东郊的工厂,承载着突破西方技术 封锁的历史使命。在"自力更生"的号召下,国营706厂、707厂也相继落成,整体上形成覆盖单晶炉、扩散炉等原始设备体系。此后数十年间,这些国营 工厂不仅培养了中国首批半导体设备专业人才,更在真空设备、晶体生长技术等领域积累了原始技术储备。 2001年,北京电子控股有限责任公司(北京电控)对上述国营资产进行战略性整合,成立七星电子(北方华创前身),主营半导体设备及精密电子元器 件业务。同年,由北京电控联合国内高校等科研机构,共同出资设立北方微电子,专注于刻蚀机、CVD(化学气相沉积)和PVD(物理气相沉积)设备 研发。这种"制造+研发"的双轨架构,标志着中国半导体设备产业从分散走向集约化发展的开端。 0 1 新中国工业化探索的深刻烙印 中国半导体设备发展史上,2001-2010年,是 ...
EDA国产化进程有望加速,信创ETF(562570)连续10天净流入
Mei Ri Jing Ji Xin Wen· 2025-05-30 06:14
截至2025年5月30日 13点57分,中证信息技术应用创新产业指数下跌1.65%。成分股方面涨跌互现,浪 潮软件领涨5.47%,天阳科技上涨4.03%,京北方上涨4.01%;开普云领跌4.61%,华大九天下跌4.12%。 信创ETF(562570)下跌1.69%,最新报价1.22元。 从资金净流入方面来看,信创ETF近10天获得连续资金净流入,最高单日获得1.48亿元净流入,合计"吸 金"1.90亿元,日均净流入达1900.63万元,体现市场投资热情。 东方证券认为,今年是十四五的收官之年,EDA 是集成电路自主可控的重点领域之一,其整合也是并 购重组的重要方向之一。今年以来两大EDA 企业连发公告,华大九天拟收购芯和、概伦电子拟收购锐 成芯微。头部EDA 企业有望持续并购重整更多优质标的,实现EDA 工具全流程替代的目标。 同时,本周海光信息、中科曙光宣布战略重组,中证信创指数与该事件相关度较高。目前。海光信息、 中科曙光分别是该指数第1、第7大权重股,合计权重超10.2%。信创ETF(562570)为跟踪该指数规模 最大的ETF。 公开信息显示,信创ETF(562570)跟踪中证信息技术应用创新产业指数 ...
科创芯片ETF(588200)近1年累计上涨超65%,近5日获资金净流入4.36亿元
Sou Hu Cai Jing· 2025-05-30 03:20
截至2025年5月30日 10:48,上证科创板芯片指数下跌1.45%。成分股方面涨跌互现,峰岹科技领涨2.11%,富创精密上涨1.36%,睿创微纳上涨1.32%;思瑞 浦领跌,寒武纪、安路科技跟跌。科创芯片ETF(588200)回调整固。拉长时间看,截至2025年5月29日,科创芯片ETF近1年累计上涨65.59%,涨幅排名可比 基金前2。 流动性方面,科创芯片ETF盘中换手2%,成交4.93亿元。拉长时间看,截至5月29日,科创芯片ETF近1年日均成交22.47亿元,居可比基金第一。 规模方面,科创芯片ETF近1周规模增长5.85亿元,实现显著增长,新增规模位居可比基金第一。份额方面,科创芯片ETF近1周份额增长2.25亿份,实现显 著增长,新增份额位居可比基金第一。 资金流入方面,拉长时间看,科创芯片ETF近5个交易日内有3日资金净流入,合计"吸金"4.36亿元。 数据显示,截至2025年4月30日,上证科创板芯片指数前十大权重股分别为中芯国际、海光信息、寒武纪、澜起科技、中微公司、芯原股份、恒玄科技、沪 硅产业、思特威、华润微,前十大权重股合计占比58.53%。 消息面上,5月27日中欧半导体上下游企 ...
早盘直击 | 今日行情关注
申万宏源证券上海北京西路营业部· 2025-05-30 02:06
周四受海外关税新增信息刺激,A 股全天放量反弹。周四 A 股结束了数周以来的缩量盘整,收出 放量的反弹阳线,个股和板块均有不错的表现。主要原因是根据新华社报道,位于纽约的美国国际贸易 法院于 5 月 28 日裁定暂停特朗普政府 4 月 2 日宣布的一揽子加征关税政策生效。受此影响,美股期指 在隔夜美股收盘后出现明显上涨,29 日亚太地区股市也出现普遍上涨。 后市展望:关税事件再次出现超预期变化,后续演进方向将对股市走势产生较大影响。5 月 28 日 美国国际贸易法院的裁定短期内已经令市场情绪明显回暖,市场开始憧憬 4 月 2 日以来的所谓"对等关 税"风波将告一段落。但不可否认的是,后续事件的演进还有一定不确定性。比如:特朗普当局是否会 对裁定进行上诉;如果上诉,在上诉过程中,4 月 2 日后宣布的所谓"对等关税"是否还会继续执行;以 及 10%所谓的"基准关税"是否会被其他企业继续提起裁定等。但不管是哪种情况,关税事件的后续影响 在继续好转是确定的事实,不确定的是好转的进程是快速好转还是在波折中好转。 热点板块:6 月可能仍是事件驱动的主题性行情,低位板块如消费、医药、调整较为充分的注科技 成长可关注。1) ...
研判2025!中国银粉银浆行业市场政策、产业链、发展现状、竞争格局及发展趋势分析:国产化替代空间仍然巨大[图]
Chan Ye Xin Xi Wang· 2025-05-30 02:02
Overview - The silver powder and silver paste industry in China is experiencing rapid growth due to the development of downstream industries such as photovoltaics and electronic information. In 2024, silver powder production is expected to reach 13,465.8 tons, and silver paste production is projected to be 22,028.9 tons. The demand for silver powder is estimated at 16,445.3 tons, while silver paste demand is expected to be 22,037.1 tons. The market size for silver powder is projected to be 49.32 billion yuan, and for silver paste, it is expected to reach 122.13 billion yuan. The "dual carbon" goals are anticipated to further boost market demand in the photovoltaic sector [1][12]. Market Policies - The Chinese government has implemented various policies to support the development of the new energy industry, which indirectly benefits the silver powder and silver paste market. Key policies include guidelines for promoting photovoltaic power generation, managing land use, and enhancing renewable energy consumption. These initiatives create a broader market space for silver powder and silver paste in the photovoltaic sector [5][7]. Industry Chain - The silver powder and silver paste industry consists of an upstream segment that includes silver mining, silver recycling, and the production of glass oxides and organic solvents. The midstream involves the manufacturing of silver powder and silver paste, while the downstream applications primarily focus on photovoltaics and electronic information sectors, with photovoltaics being the most significant market for silver powder and paste [8][10]. Development Status - The silver powder and silver paste industry is in a golden period of rapid development, driven by the booming photovoltaic and electronic information sectors. In 2024, the silver powder market size is expected to reach 49.32 billion yuan, and the silver paste market size is projected to be 122.13 billion yuan. The growth in the clean energy sector is expected to further enhance the demand for silver powder and paste [12][14]. Competitive Landscape - The domestic market for silver powder and silver paste has historically been dominated by foreign companies, but local firms are increasingly breaking this monopoly through enhanced R&D and improved product quality. Key players in the industry include Suzhou Guder, Dike, Boqian New Materials, and others, primarily located in Jiangsu, Zhejiang, and Guangdong provinces. The trend indicates a growing market share for domestic products in the global market [16][18]. Industry Representative Companies - Dike Electronic Materials Co., Ltd. has become a leading supplier of photovoltaic metallization paste, significantly contributing to the domestic market's growth. In 2024, the company's revenue is expected to reach 15.35 billion yuan, with photovoltaic conductive silver paste accounting for 83.81% of its total revenue [18]. - Suzhou Guder Electronic Co., Ltd. is recognized as a pioneer in the domestic production of photovoltaic cell conductive paste, with a comprehensive product range. The company's revenue in 2024 is projected to be 5.638 billion yuan, with its new energy materials business contributing 4.609 billion yuan [20]. Development Trends - The silver powder and silver paste industry is expected to see continuous technological innovation, focusing on developing new high-efficiency conductive materials and techniques. There will be an emphasis on green production and circular economy practices, with companies increasing investments in environmentally friendly materials and processes to reduce pollution and enhance resource recycling [22].
2025年中国CT球管行业相关政策、产业链、市场容量、竞争格局分析及发展趋势研判:需求增长、国产化加速,CT球管行业正迎来关键发展期[图]
Chan Ye Xin Xi Wang· 2025-05-30 01:56
Core Insights - The CT tube, also known as X-ray tube, is a critical component of CT equipment, directly affecting imaging quality, scanning speed, and equipment lifespan [1][9] - The cost of CT tubes accounts for approximately 20% of the total production cost of CT devices, and they have a shorter lifespan compared to the devices themselves, necessitating regular replacements [9][11] - The demand for CT tubes in China is driven by both new installations of CT devices and the maintenance of existing units, with the market capacity expected to grow significantly in the coming years [9][11] Market Capacity and Structure - In 2024, the market capacity for CT tubes in China is projected to be 35,396 units, consisting of 25,835 units from the existing market and 9,561 units from new installations [9][11] - By 2025, the market capacity is expected to reach 40,445 units, with 30,000 units from the existing market and 10,445 units from new installations [9][11] - The demand for replacement tubes due to maintenance will surpass the demand from new installations as the number of CT devices increases [9][11] Industry Development Environment - The Chinese government has implemented supportive policies for the development of high-end medical equipment and key components, including CT tubes, as part of the "14th Five-Year Plan" [4][6] - Policies encourage innovation and the enhancement of manufacturing capabilities in the medical equipment sector, particularly for CT technology [4][6] Current Industry Status - The CT equipment stock in China is expected to reach 51,670 units in 2024, with an increase of 9,561 units, and approximately 60,000 units by 2025, with an increase of 10,445 units [7] - The rapid development of the CT and medical imaging equipment industry is driven by increasing market demand and the need for domestic alternatives to imported products [7][9] Competitive Landscape - Key players in the CT tube market include companies such as United Imaging, Neusoft Medical, and others, with significant investments in R&D for domestic CT tube production [14][15] - Despite advancements in domestic production, there remains a gap between local suppliers and international manufacturers in terms of technology and manufacturing processes [14][15] Industry Trends - The CT tube industry is entering a critical development phase due to accelerated domestic production, technological upgrades, and growing demand [18] - The average lifespan of a CT tube is about 13 months, indicating a high replacement frequency during the lifespan of a CT device, which typically exceeds 10 years [11]