IT之家 10 月 21 日消息,英伟达 CEO 黄仁勋今日在德国柏林举行的 GTC 大会上宣布推出模块化 AI 工厂参考外观 DSX POD,面向以 Vera Rubin 平台为核心的完整 AI 工厂平台。
英伟达表示,该思路将计算、网络、布线、机架供电、液冷管路安装及能力验证等工作提前转移至工厂完成,计划于 2027 年初借助合作伙伴在全球交付。
DSX POD 的核心思路是将 AI 基础设施预先集成到“模块化集装箱”中,并在发货前完成平台集成与验证。设备抵达部署现场后,客户只需将其接入已经准备就绪的电力、冷却、网络等配套设施,即可开展后续部署。
英伟达称,这种方式可以减少传统数据中心建设过程中复杂的现场施工、专业人员协调和物流工作,并降低现场安装错误及零部件交付延误带来的影响。
英伟达表示,在部署场地及配套基础设施已经准备就绪的情况下,DSX POD 可将基础设施部署周期从传统方式的约 6 个月缩短至最短 14 天。
黄仁勋表示,AI 工厂只有在开始产生 token 时才能创造价值,而不仅仅是在建设完成时。英伟达称,DSX POD 旨在缩短基础设施从现场准备就绪到开始运行生产级 AI 工作负载的时间,使运营商可以更早给出推理服务并取得收入。
合作伙伴维度,英伟达正在与戴尔科技、惠普企业、联想、技嘉、鸿海旗下工业富联、和硕、广达旗下云达科技(QCT)、纬创、施耐德电气、维谛工艺等企业合作,推动 DSX POD 的制造、集成和交付。英伟达表示,全球合作伙伴计划于 2027 年初开始交付相关 AI 工厂平台。
相关阅读:
本文来源:https://www.ithome.com/1/016/138.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
IT House October 21 news, Nvidia CEO Huang Jenxun today announced the launch of the modular AI factory reference appearance DSX pod at the GTC conference in Berlin, Germany, for the complete AI factory platform with the Vera Rubin platform as the core.
Nvidia said the idea is to move computing, networking, cabling, rack power supply, liquid cooling pipeline installation and capability verification to the factory ahead of schedule, with global delivery planned for early 2027 with partners.
The core idea of DSX pod is to pre-integrate AI infrastructure into “modular containers” and complete platform integration and validation before shipment. After the equipment arrives at the deployment site, the customer can simply connect it to the ready power, cooling, network and other supporting facilities for subsequent deployment.
According to Nvidia, this approach reduces the complexity of on-site construction, professional coordination and logistics during the construction of traditional data centers, and reduces the impact of on-site installation errors and delays in the delivery of parts.
Nvidia said that with the deployment site and supporting infrastructure ready, the DSX pod can shorten the infrastructure deployment cycle from about 6 months to a minimum of 14 days.
Huang Jenxun said that the AI factory can only create value when it starts to generate tokens, not just when the construction is completed. Nvidia said the DSX pod is designed to reduce the time it takes for infrastructure to get from site readiness to start running production-grade AI workloads, allowing operators to give inference services and generate revenue earlier.
In the partner dimension, Nvidia is working with Dell Technology, HP Enterprise, Lenovo, Gigabyte, Hon Hai's Industrial Fu Lian, Heshuo, Quanta's Cloud Technology (QCT), Wistron, Schneider Electric, Vitrix Technology and other enterprises to promote the manufacture, integration and delivery of DSX pod. Nvidia said that the global partner plans to start delivering the relevant AI factory platform in early 2027.
This article is from: https://www.ithome.com/1/016/138.htm For copyright issues, please write to: 13951431913#139.com or submit a review on the right.