英伟达发布 DGX Station for Windows:本地运行 1 万亿参数 AI 模型,最高 748GB 内存NVIDIA releases DGX Station for Windows: Run 1-trillion parameter AI models locally with up to 748GB memory


AI free token/AI免费积分

IT之家 10 月 8 日消息,太平洋时间 10 月 7 日上午 10 点(北京时间 10 月 8 日凌晨 1 点)在美国旧金山举办的发布会上,英伟达发布 DGX Station for Windows,定位为桌面级 AI 超级计算机,旨在让开发者、研究人员和数据科学团队在本地运行大规模 AI 模型,将于 2026 年第四季度上市。

英伟达发布 DGX Station for Windows:本地运行 1 万亿参数 AI 模型,最高 748GB 内存

IT之家援引博文介绍,DGX Station 搭载 NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip,配备 72 核 Grace CPU、Blackwell Ultra GPU,以及最高 748GB 统一内存。

其中涵盖 252GB HBM3e GPU 内存和 496GB LPDDR5X CPU 内存,AI 算力最高可达 20 PFLOPS,兼容运行参数规模最高约 1 万亿的模型。

英伟达发布 DGX Station for Windows:本地运行 1 万亿参数 AI 模型,最高 748GB 内存

与传统高表现工作站相比,DGX Station 的核心卖点在于将数据中心级硬件、统一内存和 NVIDIA AI 软件栈压缩到桌面设备中。开发者可以在本地开展模型微调、推理、智能体开发和生成式 AI 应用测试,减少数据上传云端带来的延迟、成本和隐私压力。

NVIDIA 官方还将 DGX Station 与 Windows 生态结合,意味着开发者可在熟悉的 Windows 环境中使用本地大模型能力,并探索 AI 智能体、Copilot 和混合智能应用。

其最高功耗达到 1,600W,并需要 20A 电路,表明出这款设备虽然使用桌面形态,但更接近一台面向专业使用者的本地 AI 基础设施。

英伟达发布 DGX Station for Windows:本地运行 1 万亿参数 AI 模型,最高 748GB 内存

微软 Surface Laptop Ultra 发布会专题

本文来源:https://www.ithome.com/1/010/313.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

IT Home, October 8 — At a press conference held in San Francisco, USA on October 7 at 10:00 AM Pacific Time (1:00 AM Beijing Time on October 8), NVIDIA released DGX Station for Windows, positioned as a desktop-level AI supercomputer designed to allow developers, researchers, and data science teams to run large-scale AI models locally, with a release scheduled for Q4 2026.

NVIDIA releases DGX Station for Windows: Run 1-trillion parameter AI models locally with up to 748GB memory

According to the blog post cited by IT Home, DGX Station is powered by the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, featuring a 72-core Grace CPU, Blackwell Ultra GPU, and up to 748GB of unified memory. This includes 252GB of HBM3e GPU memory and 496GB of LPDDR5X CPU memory, with AI computing power reaching up to 20 PFLOPS, capable of running models with parameter scales of approximately 1 trillion.

NVIDIA releases DGX Station for Windows: Run 1-trillion parameter AI models locally with up to 748GB memory

Compared to traditional high-performance workstations, the core selling point of DGX Station is compressing data center-grade hardware, unified memory, and NVIDIA AI software stacks into a desktop device. Developers can conduct model fine-tuning, inference, agent development, and generative AI application testing locally, reducing the latency, cost, and privacy concerns associated with uploading data to the cloud. NVIDIA has officially combined DGX Station with the Windows ecosystem, meaning developers can use local large model capabilities in a familiar Windows environment and explore AI agents, Copilot, and hybrid intelligent applications. With a maximum power consumption of 1,600W and requiring a 20A circuit, this device, despite its desktop form factor, is closer to a local AI infrastructure aimed at professional users.

NVIDIA releases DGX Station for Windows: Run 1-trillion parameter AI models locally with up to 748GB memory

Microsoft Surface Laptop Ultra Launch Event Coverage

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry