然而，在游戏 GPU 中采用芯粒设计所面临的最大限制是更高的延迟，因为帧渲染无法容忍远距离的数据跳跃。为了解决 这个问题，AMD 必须提出一种尽可能缩短数据与计算之间距离的解决方案。根据视频中披露的一项新专利显示，AMD 可 能已经破解了"多芯粒"游戏 GPU 的设计之道。有趣的是，这项专利披露的细节是关于 CPU 的，而非 GPU，但文本和机 制都表明，它面向的是图形处理的使用场景。 那么，AMD 究竟将如何在 GPU 中使用多芯粒设计呢？根据专利的说法，其关键技术是一种"带有智能交换器的数据结构 电路"，用于连接计算芯粒与内存控制器之间的通信。这基本上就是 AMD 的 Infinity Fabric，但缩小并针对消费级 GPU 进行了优化，因为 AMD 无法使用 HBM 内存芯片。该交换器旨在优化内存访问，首先会比较图形任务的请求是否需要任 务迁移或数据复制，其决策延迟在纳秒级别。 公众号记得加星标⭐️，第一时间看推送不会错过。 来源：内容来自 wccftech 。 AMD 对未来消费级 GPU 领域有着宏大的计划，而且并非普通方案。根据传闻和最新的专利，公司预计很快将采用多芯粒 GPU。 多芯粒模块（M ...

AMD's AI Strategy and Product Deployment - AMD is focusing on customer satisfaction and large-scale deployments of its ROCm and Instinct platforms [1][2][3] - AMD highlights that 7 out of the 10 largest AI companies are using Instinct, marking significant progress since 2023 [3] - AMD emphasizes long-term investment in the Instinct platform, which is now ready for business [4] - AMD showcases rapid deployment capabilities, with customers going from initial engagement to scaled deployment in under 90 days [5] MI300 Series and Open Source Ecosystem - AMD reiterates the leadership performance and cost efficiency of the MI300 series, emphasizing its fully open-source software design [6] - AMD highlights the importance of the open-source ecosystem, noting its faster progress compared to proprietary frameworks [7][8] - AMD launched MI350 with immediate deployment and software availability, indicating product maturity [10] ROCm Software and Enterprise AI - ROCm 7 accelerates AI innovation with features like serving optimization kernels and communication libraries, supporting various data types [11] - AMD's open-source serving frameworks achieve 1.3x performance compared to B200 versus MI350 [12] - AMD is extending ROCm to make it enterprise-ready, focusing on operations and cluster management platforms [16] - AMD provides developer cloud access with GPU credits to facilitate prototyping and access to Instinct GPUs [19][20]

Core Viewpoint - HSBC believes that AMD's newly launched MI350 series AI accelerators are now competitive with NVIDIA's Blackwell platform in terms of performance and pricing strategy [2][3] Group 1: AMD's Product Launch and Market Position - AMD's stock price increased by over 4% following HSBC's upgrade of its rating from "Hold" to "Buy" and doubling the target price to $200, which is nearly 40% above the current stock price [2] - The MI350 series allows AMD to "catch up" with NVIDIA, as it competes effectively with NVIDIA's current Blackwell AI platform [2] - The upcoming MI400 series is expected to provide greater growth potential in 2026 [2] Group 2: Financial Projections and Market Expectations - Analysts predict significant upside potential for AMD's AI revenue in fiscal year 2026, estimating AI GPU revenue to reach $15.1 billion, which is 57% higher than the market's average expectation of $9.6 billion [3] - The average selling price of the MI355 accelerator is projected to be $25,000, which is $10,000 higher than previous estimates, yet still about 30% lower than NVIDIA's products, contributing to the potential success of the MI350 series [3] - If growth expectations are met, AMD could achieve a higher valuation multiple, with a target price of $200 indicating nearly 40% upside from current levels [3]

Core Insights - Marvell Technology is advancing its semiconductor technology by utilizing 5nm and 3nm CMOS technologies and is transitioning to 2nm and below, incorporating innovations such as gate-all-around transistors and backside power delivery [1] - The company is enhancing its packaging capabilities with modular redistribution layer (RDL) interposer technology, which allows for the development of larger multi-die AI accelerator solutions [2][3] - Marvell anticipates significant growth in the data center semiconductor market, projecting a total addressable market of $94 billion by 2028, with accelerated custom compute offerings expected to reach $55.4 billion at a CAGR of 53% from 2023 to 2028 [4][10] Technology and Innovation - The RDL interposer technology provides shorter die-to-die connections, reducing latency and improving power efficiency, while also allowing for the seamless replacement of defective dies, which lowers costs and increases yields [3] - This technology is compatible with HBM3/3E and XPU-type chips, and is expected to be qualified for the upcoming HBM-4 [3] Competitive Landscape - Competitors such as Broadcom and Advanced Micro Devices are also active in the custom silicon solutions and AI accelerator space, with Broadcom's semiconductor segment growing 11% year over year [5][6] - Despite the competition, Marvell's advancements in 2.5D packaging and modular RDL interposer technology position it favorably in the market [7] Financial Performance and Valuation - Marvell's shares have decreased by 34.5% year to date, contrasting with the 15% growth of the Electronics - Semiconductors industry [8] - The company trades at a forward price-to-sales ratio of 7.02X, which is lower than the industry's average of 8.72X [12] - Earnings estimates for fiscal 2026 and 2027 indicate year-over-year growth of 77.7% and 27.73%, respectively, with recent upward revisions [13]

We’re redefining what’s possible for the AI data center with a full-stack, open, and integrated platform. Our 5th Gen AMD EPYC processors deliver the performance and efficiency to meet the demands of modern AI workloads. Combined with Instinct GPUs, Pensando DPUs, and AI NICs, we’re powering seamless, scalable infrastructure. And with continued software investment, we’re helping customers deploy and scale AI faster than ever. #AdvancingAI *** Subscribe: https://bit.ly/Subscribe_to_AMD Join the AMD Red Team ...

Core Insights - NVIDIA Corporation is experiencing significant sales losses due to U.S. export restrictions, particularly from the trade conflict with China, resulting in a $2.5 billion loss in Q1 of fiscal 2026 and an anticipated $8 billion loss in Q2 [1][10] - The company is focusing on sovereign AI projects to mitigate risks associated with the trade war, establishing partnerships and infrastructure in various regions [1][5] Group 1: Sovereign AI Initiatives - Sovereign AI involves countries developing their own AI infrastructure to protect data and maintain control, with NVIDIA playing a crucial role in these efforts [2] - NVIDIA has announced major AI factory projects in Saudi Arabia, planning to deliver 18,000 AI chips, and has received strong support for its initiatives in Europe from leaders in France and Germany [2][3] - The European Union is investing $20 billion to build four AI gigafactories, and NVIDIA is collaborating with firms like Foxconn and local governments in Taiwan and the UAE to enhance AI infrastructure [3] Group 2: Competitive Landscape - Competitors like Advanced Micro Devices (AMD) and Intel are also trying to expand in the AI infrastructure space but are lagging behind NVIDIA in sovereign AI projects [6] - AMD is gaining traction with its Instinct MI300X chips, but lacks a full-stack solution comparable to NVIDIA's offerings [7] - Intel is promoting its Gaudi 3 AI chips as a cost-effective option, but its AI ecosystem is still developing and does not provide a complete package [8] Group 3: Financial Performance and Valuation - NVIDIA's shares have increased by approximately 19.1% year-to-date, outperforming the Zacks Computer and Technology sector, which gained 7% [9] - The company trades at a forward price-to-earnings ratio of 33.13, higher than the sector average of 27.37 [11] - The Zacks Consensus Estimate for NVIDIA's fiscal 2026 and 2027 earnings indicates a year-over-year increase of about 41.8% and 31.9%, respectively, with recent upward revisions for fiscal 2027 estimates [12]

作 者丨 LL Insights 编译 | 华尔街大事件 这种新趋势出现的时机，恰逢 AI 加速器供应结构性短缺。由于 HBM 产能紧张，英伟达的订单 排期已到 2026 年之后，云服务商急需一个可靠的备选供应商。AMD 的小芯片（chiplet）策略 使其能以更低的边际硅成本整合更多内存通道，相比单片设计的竞争对手拥有物料成本优势 —— 在 192GB 容量下，每块 GPU 能节省数万美元。当数据中心的电力和空间预算有限时，这种成本 差异在数万个节点的规模下会变得无法忽视。 软件方面的障碍也不再难以逾越。ROCm 6.2 版本新增了原生 vLLM 和 Bits-and-Bytes 支持、 FP8 内核及全新性能分析工具，在训练和推理场景下与 CUDA 的效率差距大幅缩小。Meta 公开 表示，其 Llama 3.1 的生产流量已完全由 MI300X 集群承载，这说明 AMD 的软件栈已经能胜任 最严苛的实时推理工作负载。随着越来越多开源仓库接受 HIP 代码提交，客户的切换成本会逐季 降低。 Instinct MI300X 是首款将 192 GB HBM3e 与 5.3 TB/s 带宽集成于一体的数据中心 ...

Core Insights - Advanced Micro Devices (AMD) shares have increased by 11.4% year-to-date, outperforming the Zacks Computer and Technology sector's growth of 7.9% and Intel Corporation's (INTC) 9.5% increase [1][8] Group 1: Financial Performance - AMD's data center revenues surged by 57.2% year-over-year to $3.674 billion in Q1 2025, accounting for 49.4% of total revenues [4][8] - AMD expects Q2 2025 revenues of $7.4 billion (+/- $300 million), indicating a year-over-year growth of approximately 27% [13] - The Zacks Consensus Estimate for AMD's Q2 2025 earnings is 54 cents per share, reflecting a year-over-year decline of 21.74% [14] Group 2: Product and Market Developments - AMD is expanding its EPYC portfolio, with Nokia adopting its fifth-gen EPYC 9005 Series processors for enhanced performance in cloud applications [5][8] - The introduction of the Instinct MI350 Series GPUs at the 2025 Advancing AI event showcases AMD's commitment to the AI market, with 4x generational AI compute gains [6][9] - Strategic partnerships with companies like Meta Platforms and OpenAI are enhancing AMD's role in AI workloads [9] Group 3: Competitive Landscape - AMD faces significant competition from NVIDIA, particularly in the data center and AI chip markets, where NVIDIA's data center revenues increased by 73.3% year-over-year [10][11] - Export restrictions on GPUs to China are expected to negatively impact AMD's revenues by approximately $700 million in Q2 2025 [12][13] Group 4: Valuation and Investment Outlook - AMD stock is currently considered overvalued, with a forward 12-month Price/Sales ratio of 6.33X compared to the industry average of 3.83X [15] - The company is advised to hold its stock for now, as near-term prospects are affected by macroeconomic uncertainties and competitive pressures [19][20]

公众号记得加星标⭐️，第一时间看推送不会错过。 来源：内容来自雅虎财经 。 毛利率下降并非Marvell股票目前面临的唯一挑战。该公司还面临着来自博通AVGO和AMD在定制硅 片解决方案领域的巨大竞争压力。 博通先进的 3.5D XDSiP 封装平台专为提升定制 AI XPU 的性能和效率而设计。博通的半导体部门 （包括其定制硅片解决方案）在 2025 财年第一季度同比增长 11%。 AMD的定制硅片解决方案和AI加速器（例如 Instinct Accelerators）为众多数据中心提供支持。 AMD的可重构Alveo Adaptable Accelerator Cards 用于加速数据中心的计算密集型应用程序。 Ma r v e ll的价格表现、估值和估值 今年迄今为止，Marvell 的股价已下跌 32.2%，而电子半导体行业则增长了 13.3%。 Marvell的营收正在增长，这得益于其定制 AI 芯片在超大规模数据中心运营商中广受欢迎。然而， Marvell Technology 的毛利率在过去几个季度波动较大，且长期呈下降趋势。 Marvell强劲的营收增长得益于其定制 AI XPU、电光解决方案 ...

Key Takeaways MU's HBM revenue topped $1B in Q2 FY25, boosted by AI and data center demand. MU's 2025 EPS estimate rose 12.1% to $7.77, signaling 497.69% year-over-year growth. AMD's Data Center revenues jumped 57.2% in Q1 2025, driven by EPYC and Instinct GPU demand.Advanced Micro Devices (AMD) and Micron Technology (MU) are major players in the semiconductor industry. AMD is primarily known for producing CPUs, GPUs, and other integrated circuits, while Micron Technology specializes in memory and storage ...