AI训练智算卡

Search documents
龙虎榜复盘 | 国产芯片大爆发,RWA持续活跃
Xuan Gu Bao· 2025-08-22 10:58
| 上榜热股 | 实时涨跌幅 ◆ 买/卖家数 ◆ | | | --- | --- | --- | | 中兴通讯 000063.SZ | +9.38% | 2/1 | | 3日 | | | | 昆仑万维 300418.SZ | +20.01% | 2/1 | | 和而泰 002402.SZ | +10.02% | 2/0 | A股稀缺AI芯片厂商 龙虎榜机构热股 今天机构龙虎榜上榜32只个股,净买入16只,净卖出16只。当日机构买入最多的个股前三位是:中兴通讯(11.05亿)、昆仑万维(2.74亿)、和而泰(2.62 亿) 和而泰 龙虎榜显示,今日2家机构净买入2.62亿。 公司目前持有摩尔线程1.244%股份。 一、国产芯片 寒武纪 个股龙虎榜 海光信息 个股龙虎榜 国产CPU、GPU龙头,国内稀缺的DCU设计厂商,公司拟吸收合并中科曙光 8月21日,DeepSeek表示最新的V3.1版本使用了UE8M0 FP8 Scale参数精度,同时在留言里透露,UE8M0 FP8是针对即将发布的下一代国产芯片设计。 上海证券认为,国产GPU虽然还未到对标国外产品的水平,但已经能够满足部分对算力有一定要求的应用场景,目前 ...
DeepSeek-V3.1正式发布,科创芯片ETF基金(588290)涨超7%,寒武纪-U涨超10%,机构:高端AI芯片国产化势在必行
2 1 Shi Ji Jing Ji Bao Dao· 2025-08-22 03:36
华鑫证券指出,高端AI芯片国产化势在必行。国产AI芯片方面,除了寒武纪、海光信息以及华为等厂 商,其余国产AI芯片厂商也呈现加速渗透的态势。摩尔线程能够提供AI训练智算卡、AI推理卡及夸娥 (KUAE)智算集群等。其最新一代GPU采用"平湖"架构,增加了FP8精度支持,大幅提升AI算力,能 够全面支撑面向DeepSeek类前沿大模型预训练的万卡集群智算中心解决方案。公司推出KUAE2集群方 案,集群内部有10240个GPU,单GPU互联达到800GB/s,性能表现良好。沐曦能够提供训推一体的C系 列GPU,该GPU具有强大的大规模并行计算能力,已在多个智算中心中获得规模化应用。此外,沐曦还 能提供智算推理为主的N系列GPU,主要面向生成式人工智能下的云端人工智能推理场景,拥有强大的 多精度混合算力。 (本文机构观点来自持牌证券机构,不构成任何投资建议,亦不代表平台观点,请投资人独立判断和决 策。) 今日,上证科创板芯片指数盘中表现强劲,涨幅一度高达7.33%。相关ETF方面,科创芯片ETF基金 (588290)盘中大涨,涨幅一度达到7.15%,成交额突破2亿元;据Wind金融终端,截至8月21日其年内 涨幅超 ...
DeepSeek正式发布新模型,还透露国产AI芯片关键信息
Xuan Gu Bao· 2025-08-21 23:22
8月21日,DeepSeek表示最新的V3.1版本使用了UE8M0 FP8 Scale参数精度,同时在留言里透露, UE8M0 FP8是针对即将发布的下一代国产芯片设计。 资料显示,FP8是当前AI计算的一种前沿低精度格式,在支持原生FP8的GPU上,大语言模型训练采用 FP8混合精度可大幅提高GPU算力,显著降低显存占用,在2022年首次被英伟达、Arm和英特尔联合提 出, A股算力芯片龙头:寒武纪、海光信息。 华为昇腾产业链:拓维信息、神州数码、华丰科技等。 此外,据官微介绍,本次升级后将支持混合推理架构,拥有更高的思考效率和更强的Agent能力。通过 Post-Training优化,新模型在工具使用与智能体任务中的表现有较大提升。 对于AI芯片,上海证券认为,国产GPU虽然还未到对标国外产品的水平,但已经能够满足部分对算力 有一定要求的应用场景,目前国产GPU进入了快速发展阶段,将从"可用"阶段进入"好用"阶段。 除了寒武纪、海光信息以及华为等厂商,华鑫证券表示,其他国产AI芯片厂商也呈现加速渗透的态 势。如摩尔线程能够提供AI训练智算卡、AI推理卡及夸娥(KUAE)智算集群等。其最新一代GPU采 用" ...
【IPO一线】摩尔线程科创板IPO获受理 募资80亿元投建图形芯片/AI训推一体芯片等项目
Ju Chao Zi Xun· 2025-06-30 13:19
Core Viewpoint - The company, Moore Threads, has officially submitted its IPO application to the Shanghai Stock Exchange, aiming to raise 8 billion yuan for the development of next-generation AI and graphics chips, enhancing its competitive edge in the GPU market [1][4]. Group 1: Company Overview - Moore Threads specializes in the research, design, and sales of GPUs and related products, focusing on high-performance computing areas such as AI, digital twins, and scientific computing since its establishment in 2020 [1]. - The company has successfully launched four generations of GPU architectures, creating a diverse product matrix that covers various applications including AI intelligence, high-performance computing, graphics rendering, and personal entertainment tools [1][2]. Group 2: Product Lines - The AI computing product line includes AI training and inference cards, AI supernode servers, and KUAE computing clusters, designed to meet the needs of AI computing centers and cloud service platforms [2]. - Professional graphics acceleration products are aimed at high-end applications such as industrial design and video editing, featuring various GPU series and integrated solutions [3]. - Desktop graphics acceleration products cater to AI PCs, gaming PCs, and office PCs, supporting multiple operating systems and CPU platforms [3]. - Intelligent SoC products are designed for AI PCs, edge intelligence, and smart vehicles, providing integrated solutions for AI training and inference needs [3]. Group 3: Fundraising and Strategic Goals - The IPO aims to raise 8 billion yuan for projects focused on the development of next-generation AI training and inference chips, graphics chips, and AI SoC chips, as well as to supplement working capital [4]. - The fundraising projects are aligned with the company's development plans, aimed at enhancing GPU performance, driving product innovation, and expanding the company's business scale [4]. - Moore Threads aims to become a leading GPU enterprise with international competitiveness, providing robust AI computing support for digital transformation across various industries [4].