戴尔科技推出高密度计算液冷基础架构，AI工厂引领数据中心革新

2024年10月16日 09:45 中关村在线

戴尔科技集团宣布推出全新集成式可扩展机架系统Dell Integrated Rack 7000（IR7000）与Dell PowerEdge M7725服务器，并为Dell PowerScale存储和Dell DataLakehouse数据管理产品再添新功能，持续巩固戴尔科技AI工厂创新成果，为大规模高密度计算及人工智能（AI）工作负载提供动力。

戴尔科技集团信息基础架构解决方案事业部总裁Arthur Lewis表示：“当今的数据中心已无法匹配多样化的AI需求，企业亟需一类可驾驭高密度计算并拥有液体冷却创新的新型系统，辅以模块化且灵活高效的设计，为其在瞬息万变的AI领域提供保持持续竞争力所需的性能支持。”

领先冷却创新技术助力加速计算

Dell IR7000以卓越的密度、更可持续的电源管理和先进的冷却技术，满足各类加速计算需求。基于开放计算项目（OCP）标准，Dell IR7000机架是大规模部署的理想之选，其设计面向未来可沿用多代，并充分适配异构技术环境。Dell IR7000的关键优势包括：

更高密度：21英寸设计以支持业界领先的CPU和GPU密度。

未来就绪，效率更进一步：机架具有更宽、更高的服务器托架，可容纳更大的CPU与GPU架构，并专为液冷设计，即使面对高达480千瓦的服务器也游刃有余，能够捕获几乎100%的热量。

更多灵活选择：Dell IR7000可同时支持戴尔科技自有网络设备和其他品牌网络设备，为客户提供更多选择、更大灵活性。

简易部署，能效更佳：利用戴尔科技的集成机架可扩展系统（Dell IRSS），IR7000的部署简单且更加节能。IRSS提供创新的机架级可扩展基础架构，针对AI工作负载进行了优化，通过完全集成的即插即用机架级可扩展系统实现了丝滑高效的安装过程。

Dell Integrated Rack 7000（IR7000）

戴尔科技全新AI就绪平台，专为Dell IR7000打造

全新Dell PowerEdge M7725拥有强大的高性能计算能力，轻松应对计算密集型工作负载，是研究机构、金融技术行业和高等教育等多种环境的理想选择。PowerEdge M7725专为IR7000打造，提供更高计算能力，可扩展至每机架24000-27000核心，支持64或72个双路节点，搭载第五代AMD EPYC处理器，其前置I/O接口更助力实现高速IO连接，为高需求应用提供无缝连接。通过采取CPU直接液冷（DLC）技术以及同集成机架快速连接而实现的风冷技术，PowerEdge M7725以其节能型外形尺寸助力更加可持续的部署。

面向AI时代的非结构化存储和数据管理创新

Dell PowerScale迎来重磅升级，标志着戴尔科技在非结构化数据存储产品组合上的又一轮创新，进一步提升AI应用性能，并提供简化的全球数据管理功能。

作为全球率先通过NVIDIA DGX SuperPOD验证的以太网存储解决方案，升级后的Dell PowerScale加强了数据管理策略、提高工作负载性能并为AI工作负载提供了更多支持。

广告
挖掘更多数据：PowerScale元数据和现代化数据湖仓Dell Data Lakehouse，可释放数据洞察，实现更快速智能的决策。即将推出的用于NVIDIA NeMo服务与检索增强生成（RAG）框架的戴尔科技开源文档加载器旨在帮助客户缩短数据摄取时间，降低计算和GPU成本。

实现更密集存储：全新61TB硬盘提高了容量和效率，数据中心的存储空间占用因此减少了一半；得益于全新硬盘，客户现在可以通过在更大数据集上训练AI模型来对其进行微调。

提升AI性能：通过前端NVIDIA InfiniBand功能和200GbE以太网适配器支持，AI工作负载性能得到增强，吞吐量可提升高达63%。

增强后的Dell Data Lakehouse数据管理平台，可为客户节省时间，并通过灾难恢复（容灾）、自动模式发现、综合API管理和自助式全栈升级等新功能改善企业运营。

上市时间

Dell IR7000将于2025年第一季度在全球上市。

Dell PowerEdge M7725将于2025年第一季度在全球上市。

Dell PowerScale升级功能将于2024年第四季度在全球上市。

Dell Data Lakehouse升级功能将于2025年上半年在全球上市。

Dell Announces Integrated Rack-Scalable Systems For Dell AI Factory
written by Harold Fritts October 15, 2024

Dell Technologies has unveiled a series of integrated rack-scalable systems, including server, storage, and data management innovations designed for Dell AI Factory. These systems are expected to enhance high-density computing and AI workloads on a large scale.

Arthur Lewis, President of Dell Technologies’ Infrastructure Solutions Group, emphasized the growing demand for AI in data centers, noting that current infrastructures struggle to meet these requirements. He highlighted that Dell’s new systems provide the necessary performance to help organizations stay competitive in this rapidly changing AI landscape.

Cooling Innovations, Density, and Flexibility
The Dell Integrated Rack 7000 (IR7000) is designed to meet accelerated computing needs with superior density, efficient power management, and advanced cooling capabilities. As an Open Compute Project (OCP) standards-based rack, the IR7000 is engineered for large-scale deployments and future technology environments.

The Dell Integrated Rack 7000 (IR7000) is built to handle the growing demands of accelerated computing. It offers a combination of superior density, sustainable power management, and advanced cooling technologies. As a rack adhering to Open Compute Project (OCP) standards, the IR7000 is optimized for large-scale deployments. It can accommodate future technology across multiple generations of hardware and diverse environments.

The IR7000 is designed with density in mind. Its 21-inch frame is engineered to support industry-leading CPU and GPU configurations, maximizing the use of space to handle high-performance computing tasks. This ensures that the rack can support the most demanding workloads, including AI and machine learning applications that rely on substantial processing power.

The IR7000 design includes server sleds that are wider and taller than traditional setups to accommodate current, more significant CPU and GPU architectures. These expanded dimensions enable the rack to support more powerful and sophisticated hardware, ensuring it remains relevant as technology evolves. Additionally, the rack is natively designed for liquid cooling, making it capable of cooling future deployments that generate up to 480KW of heat. The system is engineered to capture nearly 100% of produced heat, ensuring energy efficiency and sustainability.

Another key feature of the IR7000 is its flexibility. It is designed to provide users with greater choice in their technology configurations. It supports Dell’s networking solutions and off-the-shelf networking options, allowing organizations to integrate the system into various IT environments. This ensures customers are not locked into a single vendor solution, allowing for more versatile infrastructure choices.

Dell’s Integrated Rack Scalable Systems (IRSS) also simplifies deployment by offering a fully integrated, plug-and-play rack-scale infrastructure. The IRSS is optimized for AI workloads, letting organizations set up and scale their systems quickly and efficiently. This setup reduces the complexity of installation and ensures energy-efficient operation, making it an ideal solution for businesses looking to scale their AI operations without unnecessary complications.

By combining cooling, density, and flexibility, the IR7000 is positioned as a leading solution for organizations needing to support the most advanced computing environments, particularly in AI and machine learning.

AI-Ready Platforms

The Dell IR7000 supports two AI-focused platforms:

Dell PowerEdge XE9712: Designed for large language model (LLM) training and real-time inferencing at scale, this platform offers high-performance, dense acceleration. It includes NVIDIA GB200 NVL72 technology, connecting up to 36 NVIDIA Grace CPUs and 72 NVIDIA Blackwell GPUs in a rack-scale design. This configuration enables AI inference speeds up to 30 times faster than previous generations, with up to 25 times better energy efficiency than air-cooled systems.

Dell PowerEdge M7725: Built for environments such as research, government, and fintech, this system is optimized for dense compute, scaling between 24,000 and 27,000 cores per rack. It is powered by 5th Gen AMD EPYC CPUs and integrates direct liquid cooling (DLC) and air cooling for more sustainable deployments.

Unstructured Storage and Data Management Innovations

Dell Technologies’ unstructured data storage portfolio updates are designed to improve AI application performance while simplifying global data management. The Dell PowerScale, the first Ethernet storage certified for NVIDIA DGX SuperPOD, brings many improvements.

Enhanced discoverability: PowerScale metadata combined with Dell Data Lakehouse allows faster decision-making through improved data insights. An upcoming open-source document loader will improve data ingestion and reduce compute and GPU use costs.
Denser storage: New 61TB drives allow customers to train AI models on larger datasets, doubling storage capacity while reducing the data center footprint.
Boosted AI performance: AI workloads are further enhanced with support for NVIDIA InfiniBand and 200GbE Ethernet adapters, which offer up to 63% faster throughput.
Enhancements to the Dell Data Lakehouse platform include disaster recovery, automated schema discovery, comprehensive management APIs, and self-service full-stack upgrades, helping organizations save time and streamline operations. Dell also offers Optimization Services for Data Cataloging and Implementation Services for Data Pipelines to assist customers in scaling AI and business use cases with better data management.

Dell Generative AI Solutions with Intel

Dell has also introduced Generative AI Solutions as part of the Dell AI Factory, featuring preconfigured, validated platforms for various AI use cases. These solutions include the Dell PowerEdge XE9680, Intel Gaudi 3 AI accelerators, and Dell’s storage, networking, and software stack. They are designed for modern content creation, digital assistants, and code-generation workflows.

These new solutions from Dell Technologies will redefine AI infrastructure by addressing the critical needs of high-density computing, cooling efficiency, and scalable AI deployment for enterprises to remain competitive.

Product Availability

The Dell IR7000 will be available globally in Q1 CY2025.
The Dell PowerEdge XE9712 is currently available for select customers.
The Dell PowerEdge M7725 will be available globally in Q1 CY2025.
Updates to Dell PowerScale will be available in Q4 CY2024.
Dell Data Lakehouse updates are expected in the first half of CY2025.
Dell Generative AI Solutions with Intel will be available in Q4 CY2024.

戴尔宣布为戴尔 AI 工厂推出集成式机架可扩展系统
写 哈罗德弗里茨 2024 年 10 月 15 日

戴尔科技集团推出了一系列集成式机架可扩展系统，包括专为以下目的而设计的服务器、存储和数据管理创新： 戴尔人工智能工厂。 这些系统有望大规模增强高密度计算和人工智能工作负载。

戴尔科技基础设施解决方案集团总裁 Arthur Lewis 强调了数据中心对人工智能的需求日益增长，并指出当前的基础设施难以满足这些要求。他强调，戴尔的新系统提供了必要的性能，可帮助组织在快速变化的人工智能领域保持竞争力。

冷却创新、密??度和灵活性

Dell Integrated Rack 7000 (IR7000) 旨在满足加速计算需求，具有卓越的密度、高效的电源管理和先进的冷却功能。作为基于开放计算项目 (OCP) 标准的机架，IR7000 专为大规模部署和未来技术环境而设计。

Dell Integrated Rack 7000 (IR7000) 专为满足日益增长的加速计算需求而设计。它结合了卓越的密度、可持续的电源管理和先进的冷却技术。作为符合开放计算项目 (OCP) 标准的机架，IR7000 针对大规模部署进行了优化。它可以适应多代硬件和各种环境的未来技术。

IR7000 的设计充分考虑了密度。其 21 英寸机架设计用于支持业界领先的 CPU 和 GPU 配置，从而最大程度地利用空间来处理高性能计算任务。这确保机架能够支持最苛刻的工作负载，包括依赖强大处理能力的 AI 和机器学习应用程序。

IR7000 设计包括比传统设置更宽更高的服务器托架，以适应当前更重要的 CPU 和 GPU 架构。这些扩展的尺寸使机架能够支持更强大、更复杂的硬件，确保其在技术发展过程中保持相关性。此外，该机架专为液体冷却而设计，使其能够冷却未来产生高达 480KW 热量的部署。该系统旨在捕获几乎 100% 的产生热量，确保能源效率和可持续性。

IR7000 的另一个关键特性是其灵活性。它旨在为用户提供更多技术配置选择。它支持戴尔的网络解决方案和现成的网络选项，使组织能够将系统集成到各种 IT 环境中。这确保客户不会被锁定在单一供应商的解决方案中，从而提供更多功能的基础架构选择。

戴尔的集成机架可扩展系统 (IRSS) 还通过提供完全集成的即插即用机架级基础设施来简化部署。IRSS 针对 AI 工作负载进行了优化，让组织能够快速高效地设置和扩展其系统。此设置降低了安装的复杂性并确保了节能运行，使其成为希望扩展 AI 运营而无需不必要复杂性的企业的理想解决方案。

通过结合冷却、密度和灵活性，IR7000 被定位为需要支持最先进计算环境（特别是在人工智能和机器学习领域）的组织的领先解决方案。

人工智能就绪平台

Dell IR7000 支持两个以 AI 为中心的平台：

戴尔 PowerEdge XE9712：该平台专为大规模语言模型 (LLM) 训练和实时推理而设计，可提供高性能、密集加速。它包括 NVIDIA GB200 NVL72 技术，在机架级设计中连接多达 36 个 NVIDIA Grace CPU 和 72 个 NVIDIA Blackwell GPU。这种配置使 AI 推理速度比前几代产品快 30 倍，能源效率比风冷系统高 25 倍。
戴尔 PowerEdge M7725：该系统专为研究、政府和金融科技等环境而设计，针对密集计算进行了优化，每个机架可扩展至 24,000 到 27,000 个核心。它由第五代 AMD EPYC CPU 提供支持，并集成直接液体冷却 (DLC) 和空气冷却，以实现更可持续的部署。

非结构化存储和数据管理创新

戴尔科技的非结构化数据存储产品组合更新旨在提高 AI 应用程序性能，同时简化全球数据管理。 戴尔 PowerScale作为首款获得 NVIDIA DGX SuperPOD 认证的以太网存储，带来了诸多改进。

增强可发现性：PowerScale 元数据与 Dell Data Lakehouse 相结合，可通过改进的数据洞察力加快决策速度。即将推出的开源文档加载器将改善数据提取并降低计算和 GPU 使用成本。

更密集的存储：新的 61TB 硬盘允许客户在更大的数据集上训练 AI 模型，从而将存储容量加倍，同时减少数据中心占用空间。
增强人工智能性能：通过支持 NVIDIA InfiniBand 和 200GbE 以太网适配器，AI 工作负载得到进一步增强，吞吐量提高高达 63%。

关于我们

北京汉深流体技术有限公司是丹佛斯中国数据中心签约代理商。产品包括FD83全流量自锁球阀接头；液冷通用快速接头UQD & UQDB；OCP ORV3盲插快换接头BMQC；EHW194 EPDM液冷软管、电磁阀、压力和温度传感器及Manifold的生产。在国家数字经济、东数西算、双碳、新基建战略的交汇点，公司聚焦组建高素质、经验丰富的液冷工程师团队，为客户提供卓越的工程设计和强大的客户服务。

公司产品涵盖：丹佛斯液冷流体连接器、EPDM软管、电磁阀、压力和温度传感器及Manifold。
未来公司发展规划：数据中心液冷基础设施解决方案厂家，具备冷量分配单元（CDU）、二次侧管路（SFN）和Manifold的专业研发设计制造能力。

- 针对机架式服务器中Manifold/节点、CDU/主回路等应用场景，提供不同口径及锁紧方式的手动和全自动快速连接器。
- 针对高可用和高密度要求的刀片式机架，可提供带浮动、自动校正不对中误差的盲插连接器。以实现狭小空间的精准对接。
- 基于OCP标准全新打造的液冷通用快速接头UQD & UQDB ;OCP ORV3盲插快换接头BMQC , 支持全球范围内的大批量交付。