Here are the contents translated into English:

NVIDIA Announcements at GTC 2025

  1. AIQ: NVIDIA emphasizes a mechanism for observability and transparency, enabling development teams to monitor Agent activity in real-time and continuously optimize systems based on performance data.

  2. Upgraded Cosmos Model: NVIDIA unveiled an upgraded version of the Cosmos model. The original Cosmos model is capable of predicting future frames based on existing ones, generating detailed videos from text/image input data, and predicting scenario evolution by combining current states with actions.

  3. Cosmos Model Capabilities: This upgraded model includes three capabilities: Cosmos Transfer, which converts structured video text input to controlled real-world video output; Cosmos Predict, which generates virtual world states from multimodal inputs, supporting multi-frame generation and action trajectory prediction; and Cosmos Reason, an open and fully customizable model with spatiotemporal awareness, capable of understanding video data through thought chains and predicting interaction results.

  4. Isaac GR00T N1 Humanoid Robot Base Model: NVIDIA also trained the Isaac GR00T N1 humanoid robot base model, which employs a dual-system architecture featuring fast-reacting “System 1” and deep reasoning “System 2”. This model can handle general tasks like grasping, moving, and operating arms, and can be fully customized for specific robots.

  5. NVIDIA’s Three-in-One Computing Ecosystem: NVIDIA has built a three-in-one computing ecosystem, which was emphasized by Huang Jia last year at GTC: DGX (large GPU servers), AGX (embedded platforms for edge computing and autonomous systems), and Omniverse+Cosmos (data generation computers). This system can give birth to trillions of robots.

In summary, this article focuses on NVIDIA’s announcements at the GTC 2025 conference, including an upgraded Cosmos model, Isaac GR00T N1 humanoid robot base model, and NVIDIA’s three-in-one computing ecosystem.

Translation

这篇文章主要介绍了NVIDIA在2025年的GTC大会中宣布的一些新产品和技术。以下是文章的主要内容：

  1. AIQ: NVIDIA强调可观察性和透明度机制，这使开发团队能够实时监控Agent的活动，并且基于性能数据持续优化系统。

  2. Cosmos模型升级版: NVIDIA公布了Cosmos模型的升级版。Cosmos是一个能通过现有画面预测未来画面的模型，它可以从文本/图像输入数据生成详细的视频，并且通过将当前状态与动作相结合来预测场景的演变。

  3. Cosmos模型包含三部分能力: 这次升级后的模型包含了三部分能力：Cosmos Transfer、Cosmos Predict和Cosmos Reason。Cosmos Transfer可以将结构化的视频文字输入转换为可控的真实感视频输出；Cosmos Predict能够从多模态输入生成虚拟世界状态，支持多帧生成和动作轨迹预测；Cosmos Reason是一个开放且可完全定制的模型，它具有时空感知能力，可以通过思维链推理来理解视频数据，并且预测交互结果。

  4. Isaac GR00T N1人形机器人基础模型: NVIDIA还训练了Isaac GR00T N1人形机器人基础模型，这采用双系统架构，有快速反应的“系统1”和深度推理的“系统2”。这个模型能够处理抓取、移动、双臂操作等通用任务，并且可以根据具体的机器人进行完全定制。

  5. 英伟达构建了三位一体的算力体系: NVIDIA已经构建了三位一体的算力体系，从去年开始老黄就在GTC上强调一个「三台计算机」的概念：DGX、大型GPU服务器；AGX，NVIDIA为边缘计算和自主系统设计的嵌入式计算平台；Omniverse+Cosmos，数据生成计算机。靠着这套体系，可以诞生十亿级的机器人。

总之，这篇文章重点介绍了NVIDIA在GTC2025大会上宣布的一些新产品和技术，包括升级版的Cosmos模型、Isaac GR00T N1人形机器人基础模型，以及英伟达构建的三位一体的算力体系。

Reference:

https://www.youtube.com/watch?v=4wZYrzC-pTo

<
Previous Post
An interview with Sam Altman - building a consumer tech company
>
Next Post
An opinion about MCP