证券日报网讯1月19日，沐曦股份在互动平台回答投资者提问时表示，公司训推一体GPU具备国内领先 的计算能力，拥有丰富的标量、矢量和张量计算单元，支持多精度计算，能够满足人工智能训练和推 理、通用计算、科学计算等多元化的算力需求，大幅提升了智算集群的多场景融合应用能力。同时，公 司是全球少数几家掌握了复杂多级缓存结构的GPU企业，凭借独特的缓存组网技术，能够针对不同场景 灵活自适应多级缓存控制策略，实现对高速缓存的精细管控、有效提升缓存命中率，满足GPU高算力下 对于高带宽的需求。智算推理GPU曦思N系列产品是公司于2022年推出的首款产品，定位为应用于人脸 识别、图像识别、文本分类等传统人工智能推理场景的算力芯片。公司曦思N100产品系面向传统人工 智能场景，内置性能强劲的视频处理器和运算核心，可广泛应用于智慧城市、智慧交通、智慧教育、智 能视频处理等场景；公司曦思N系列后续迭代产品(N260与N300系列(在研))系面向生成式人工智能场 景，拥有多精度混合算力、大容量显存和较高的能效比，可广泛应用于大模型推理、生成式应用等场 景。 ...

Industry Overview - The GPU product market has significant growth potential, particularly in the rapidly developing artificial intelligence (AI) computing sector [2] - The company is projected to capture approximately 1% of the domestic AI chip market share in 2024, benefiting from the accelerating trend of domestic substitution [2] Company Performance - The company has demonstrated strong technical advantages and product competitiveness, leading to widespread application of its products across major downstream industries in AI [2] - There is a continuous repurchase from existing customers and ongoing acquisition of new clients, contributing to rapid revenue growth [2] - The company has a robust order backlog and ample potential business opportunities, indicating a healthy growth trajectory [2] Business Strategy - The company is focusing on various types of computing power base customers while actively expanding into internet and telecommunications sectors, which has started to yield positive results [2] - The internet and telecommunications sectors are expected to become significant growth drivers for the company's future business [2]

（编辑 任世碧） 证券日报网讯 1月19日，沐曦股份在互动平台回答投资者提问时表示，受益于加速国产替代的行业趋 势，公司未来业务增长空间较大。从公司自身来看，凭借突出的技术优势和较强的产品竞争力，沐曦股 份产品已在人工智能下游各主要行业得到广泛应用，存量客户持续复购、新客户不断开拓，公司收入规 模呈快速增长，目前在手订单及潜在商机充足；公司以各类型算力底座客户为基本盘，同时大力开拓互 联网和运营商客户并已初见成效，未来亦将成为重要的业务增长极，提供发展动能。 ...

Core Viewpoint - The company, Muxi Co., has demonstrated its industry-leading capabilities in high bandwidth and multi-card interconnection, positioning itself as a key player in the AI computing sector [2] Group 1: Technology and Innovation - The company has developed the MetaXLink high-speed interconnection technology, which overcomes traditional PCIe bus limitations in bandwidth and latency, achieving interconnection bandwidth performance comparable to H200 [2] - The GPU products of the company can support various interconnection topologies and supernode architectures, catering to diverse needs from small to large-scale training [2] - The company has effectively balanced computing power, interconnection communication, and software ecosystem, enabling its GPU products to perform robustly in actual intelligent computing cluster operations [2] Group 2: Market Position and Applications - The company's GPU products are capable of executing massive parameter model distributed training tasks stably and without interruption, providing critical technical support for complex AI tasks [2] - The products support the construction of heterogeneous fusion computing clusters, allowing for elastic expansion and helping downstream customers optimize existing resources during the domestic substitution process [2] - The solutions provided by the company enhance the overall computing power efficiency and offer optimization plans for expanding existing intelligent computing clusters [2]

Core Viewpoint - The company has established a stable supply chain system and has accumulated rich experience in supply chain management, ensuring high-quality and sustainable supply and delivery of products [2] Group 1: Supply Chain Management - The company possesses a dynamic and flexible inventory management mechanism [2] - It has initiated the layout of a domestic supply chain early and has planned and designed products based on this domestic supply chain [2] - The company has made certain breakthroughs in the localization of key processes such as wafer production, packaging and testing, storage particles, EDA, and IP [2] Group 2: Market Position and Strategy - As the company's market influence and industry chain resources gradually increase, it will continue to cultivate domestic suppliers [2] - The company aims to further promote the localization of key supply links, enhancing the competitiveness of the domestic integrated circuit industry chain [2]

证券日报网讯1月19日，沐曦股份在互动平台回答投资者提问时表示，公司目前正在大力开拓互联网企 业和运营商两类关键客户，稳步推进相关产品测试、导入及商务洽谈，并已取得一定成效。 ...

证券日报网讯1月19日，沐曦股份在互动平台回答投资者提问时表示，沐曦股份是国内少数真正实现千 卡集群大规模商业化应用的GPU供应商，并正在研发和推动万卡集群的落地，目前已成功支持 128BMoE大模型等完成全量预训练。公司与整机服务器、操作系统、运维管理平台、主流AI框架、主 流大模型等上下游生态广泛适配，通过芯片层、框架层、模型层的深度协同优化，为头部大模型分布式 推理提供千卡高性能国产算力，助力"国产算力+大模型产业"实现从技术突破到商业化落地的跨越式发 展。 ...

格隆汇1月19日｜有投资者在互动平台向沐曦股份提问：公司在互联网大厂客户拓展方面进展如何？沐 曦股份回复称，公司目前正在大力开拓互联网企业和运营商两类关键客户，稳步推进相关产品测试、导 入及商务洽谈，并已取得一定成效。 请注意投资风险！ ...

Core Viewpoint - The company has made significant technological advancements with its next-generation C700 series, which features a fully domestic supply chain and enhanced performance metrics compared to its current products [1] Group 1: Product Development - The C700 series utilizes advanced domestic manufacturing processes, establishing a closed-loop supply chain from design to packaging and testing [1] - The core technology of the C700 series is independently controllable, indicating a strong focus on self-sufficiency in technology [1] Group 2: Performance Metrics - The performance of the C700 series is comparable to the H100, with substantial improvements in computing power, storage capacity, communication capabilities, and energy efficiency compared to existing products [1]

Core Viewpoint - The company has made significant technological advancements with its next-generation C700 series, which features a fully domestic supply chain and enhanced performance metrics compared to its current products [1] Group 1: Product Development - The C700 series utilizes advanced domestic manufacturing processes, establishing a closed-loop supply chain from design to packaging and testing [1] - The core technologies of the C700 series are independently controllable, indicating a focus on self-sufficiency in technology [1] Group 2: Performance Metrics - The performance of the C700 series is comparable to the H100, with substantial improvements in computing power, storage capacity, communication capabilities, and energy efficiency compared to existing products [1]