内容来源：https://aibusiness.com/data-centers/intel-new-gpu-ai-portfolio

内容总结：

英特尔发布新一代数据中心GPU 抢占AI推理芯片市场新高地

在2025年OCP全球峰会上，英特尔正式推出代号“新月岛”的全新数据中心GPU，该芯片搭载160GB LPDDR5X内存，专为大规模AI推理工作负载设计。这是继去年高迪3芯片市场反响平平后，英特尔在AI芯片领域的又一次重要尝试。

据悉，这款基于Xe3P微架构的芯片主打能效优势，旨在支撑大语言模型等复杂AI任务。英特尔首席技术官萨钦·卡蒂指出，AI产业正从静态训练转向实时推理，而“新月岛”将为激增的令牌服务提供关键算力支持。

当前英特尔正面临严峻挑战。财报显示该公司在2024年亏损高达190亿美元，今年第一季度再亏8.21亿美元，并于7月宣布大规模裁员。为扭转局面，英特尔先后获得美国政府89亿美元股权投资及英伟达50亿美元战略投资，双方将联合开发整合CPU与GPU技术的系统级芯片。

按计划，该芯片将于2026年下半年开始送样测试。在全球AI芯片竞赛白热化的背景下，这款新产品的市场表现将对英特尔的转型之路产生重要影响。

中文翻译：

由谷歌云赞助
如何选择首个生成式AI应用场景
开展生成式AI应用时，应首先关注能够提升人类信息交互体验的领域。

这家芯片制造商推出了代号"新月岛"的新型AI数据中心GPU，配备160GB显存，专为推理工作负载设计。
英特尔正以代号"新月岛"的新型数据中心GPU再次尝试在AI芯片竞赛中站稳脚跟。
这款即将加入英特尔AI加速器产品组合的新品，将承接去年发布的Gaudi 3芯片——该产品未能引发显著市场反响。
在加州圣何塞举行的2025 OCP全球峰会上发布的"新月岛"，据英特尔宣称可提供"高性能内存容量与能效表现"，以满足日益增长的AI推理工作负载需求。
基于英特尔Xe3P微架构并优化能效表现，该芯片配备160GB LPDDR5X显存，可支持大语言模型及其他高要求AI工作负载。
英特尔强调还将支持令牌即服务提供商和推理应用场景，该公司预计这类工作负载将成为主流。
英特尔首席技术官萨钦·卡蒂阐释"新月岛"的意义时表示："在代理型AI推动下，人工智能正从静态训练转向实时化、泛在化的推理。扩展这些复杂工作负载需要异构系统，通过开放软件栈将合适芯片匹配到对应任务。"
"随着令牌量激增，英特尔至强架构数据中心GPU将提供客户所需的高效运行空间及更多价值。"
该芯片的客户送样计划于2026年下半年启动。
英特尔亟需"新月岛"获得成功已是不争事实。该公司2024年报告了190亿美元巨额净亏损，今年第一季度又出现8.21亿美元赤字，遂于7月宣布裁员数千人。
此后美国政府于8月同意斥资89亿美元购入普通股，获得10%股权。随后在9月，始终引领全球AI芯片与基础设施市场的英伟达宣布投资50亿美元获取约4%股权。
作为协议组成部分，双方宣布将共同开发结合英特尔CPU技术与英伟达GPU知识的系统级芯片，并合作开发PC产品。

您可能还喜欢

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
The chip maker unveiled Crescent Island, a new AI data center GPU with 160GB memory, targeting inference workloads.
Intel is making another attempt to gain traction in the AI chip race in the form of a new data center GPU codenamed Crescent Island.
The next addition to Intel's AI accelerator portfolio will follow last year's release of its Gaudi 3 chip, which has failed to make much of an impact.
Crescent Island, which was announced at the 2025 OCP Global Summit in San Jose, California, is claimed by Intel to offer "high performance memory capacity and energy-efficient performance" to meet the growing demands of AI inference workloads.
Based on Intel's Xe3P microarchitecture with optimized performance per watt, it offers 160GB of LPDDR5X memory to support large language model and other demanding AI workloads.
Intel noted that tokens-as-a-service providers and inference use cases will also be supported, which the company envisions will become the dominant type of workload.
Intel CTO Sachin Katti explained the significance of Crescent Island, stating: "AI is shifting from static training to real-time, everywhere inference -- driven by agentic AI. Scaling these complex workloads requires heterogeneous systems that match the right silicon to the right task, powered by an open software stack.
"Intel's Xe architecture data center GPU will provide the efficient headroom customers need -- and more value -- as token volumes surge."
Customer sampling of the chip is scheduled to get under way in the second half of 2026.
That Intel needs Crescent Island to be a success is not in doubt. The company reported a massive $19 billion net loss in 2024, followed by an $821 million deficit in the first quarter of this year, which ed to an announcement of thousands of job cuts in July.
Since then, the U.S. government agreed to make an $8.9 billion investment of common stock in August, giving it a 10% equity stake. Then in September, Nvidia -- which has led the way in the global market for AI chips and infrastructure -- announced it was investing $5 billion for a stake of around 4%.
As part of the deal, the pair said it will co-develop a system-on-a-chip that combines Intel's CPU expertise with Nvidia's GPU knowledge, plus develop PC products.
You May Also Like