英伟达（Nvidia）于周五推出的RTX Spark笔记本电脑专为在笔记本电脑上运行AI应用程序而设计。据英伟达称，用户无需使用命令行，即可从开机时就开始安装诸如GitHub Copilot、Hermes、OpenClaw和Perplexity等AI工具。
Nvidia’s RTX Spark laptops, which launched on Friday, are built to run AI agents on the laptop itself. Users can install agents such as GitHub Copilot, Hermes, OpenClaw and Perplexity from the first start-up, without a command line, Nvidia said.
每台笔记本电脑都搭载了一块由Blackwell RTX显卡和Grace处理器组成的超级芯片，拥有高达128GB的共享内存以及1千万亿次浮点运算（1 petaflop）的AI处理能力。这使得这些笔记本电脑能够在本机上运行大型语言模型和AI应用程序。该处理器的设计是由英伟达与联发科（MediaTek）共同完成的。
Each laptop runs on a single superchip that pairs a Blackwell RTX graphics chip with a Grace processor. It has up to 128GB of shared memory and up to 1 petaflop of AI performance. Nvidia says that lets the laptops run large language models and AI agents locally. MediaTek worked with Nvidia on the processor design.
“RTX Spark将英伟达的AI技术、RTX显卡性能以及CUDA计算能力与微软Windows操作系统结合在一起，为用户提供了用于创作、游戏以及运行个人AI应用程序的强大平台，”英伟达创始人兼首席执行官黄仁勋（Jensen Huang）表示。
“RTX Spark brings the power of NVIDIA AI, RTX graphics and CUDA together with Microsoft Windows in one PC for creating, gaming and running personal AI agents,” said Jensen Huang, Nvidia’s founder and chief executive.
“我们正在将Windows打造成为一个支持‘混合智能’的操作系统：在适合本地运行的场景下，AI可以独立运行；而在需要时，它也可以通过云端进行处理，”微软执行副总裁Yusuf Mehdi说道。
“We’re building Windows to be the platform for hybrid intelligence, where AI can run locally when it makes sense and reach the cloud when needed,” said Yusuf Mehdi, executive vice president at Microsoft.
为了支持这些AI应用程序的运行，英伟达于周四发布了Nemotron 3.5 Bolt模型。该模型拥有740亿个参数，其中90亿个参数会在运行时被激活，并被存储在24GB的内存中。该模型仅适用于支持RTX Spark技术的笔记本电脑。Nemotron 3.5 Bolt的模型权重、训练数据及相关代码均属于开源资源，且可以与多种AI框架（如Llama.cpp、Ollama、LM Studio、Unsloth和vLLM）配合使用。
A 74B model in 24GB To run those agents, Nvidia released Nemotron 3.5 Bolt on Thursday. It has 74 billion parameters, with 9 billion active at a time, squeezed into 24GB of memory. Nvidia says it runs on supported RTX Spark configurations. Its weights, training data and recipes are open, and it works with llama.cpp, Ollama, LM Studio, Unsloth and vLLM.
此外，英伟达还计划在10月23日推出一款内存容量为4GB的简化版本，主要用于语音助手和游戏角色的开发。英伟达的开源Switchyard模型路由器允许AI应用程序在本地与云端模型之间分配计算任务，从而减少对云服务的依赖。英伟达在一篇博客文章中表示，Hermes工具也将在10月底加入该产品线。
A smaller 4GB version, for voice assistants and game characters, follows on 23 October. Nvidia’s open source Switchyard model router lets agents split work between local and cloud models, to cut cloud use. Hermes will add it later in October, Nvidia said in a blog post.
另一款免费工具Nvidia PAIR则允许RTX Spark笔记本电脑将AI模型共享给家庭网络中的其他计算机（这些计算机可以运行Windows、Linux或macOS操作系统）。
Another free tool, Nvidia PAIR, lets the laptop serve AI models to other computers on a home network. They can run Windows, Linux or macOS.
通过控制智能体的行为，英伟达的 OpenShell 运行时和微软的执行容器让用户能够控制智能体的权限范围。OpenClaw 和 GitHub Copilot 已经支持微软的容器，而 Hermes 也将在 10 月底前加入 OpenShell 的控制功能。
Controlling what agents do Nvidia’s OpenShell runtime and Microsoft Execution Containers let users control what agents are allowed to do. OpenClaw and GitHub Copilot already support Microsoft’s containers, and Hermes will add OpenShell controls by the end of October.
“我们将 Hermes 带到了 RTX Spark，这样用户就能在 Windows 笔记本电脑上本地运行开源 AI 模型和智能体，从而更好地控制自己的数据，并自由地根据自身工作流程定制 AI，”Hermes 的开发者 Nous Research 首席执行官 Dillon Rolnick 表示。
“We brought Hermes to RTX Spark so people can put open AI models and agents to work locally on a Windows laptop, with greater control over their data and the freedom to shape AI around their own workflows,” said Dillon Rolnick, chief executive of Nous Research, which makes Hermes.
创作者、开发者和游戏：Adobe 已针对该芯片优化了 Photoshop、Premiere 和 Substance 3D。两家公司在周四的联合公告中表示，AI 处理、编辑、色彩和特效的速度最高可达原来的两倍。Photoshop 和 Premiere 目前已在测试版中支持该功能，而 Substance 3D 也即将推出。
Creators, developers and games Adobe has optimised Photoshop, Premiere and Substance 3D for the chip, the two companies said in a joint announcement on Thursday. They say AI processing, editing, colour and effects run up to twice as fast. Photoshop and Premiere support it in beta releases now, and Substance 3D is coming soon.
对于开发者而言，英伟达和微软发布了首个适用于 Windows on Arm 的 CUDA 版本 PyTorch 构建包。他们表示，排名前 15,000 的 Python 包中已有 97% 做好准备。
For developers, Nvidia and Microsoft list the first PyTorch build with CUDA for Windows on Arm. They say 97% of the top 15,000 Python packages are ready.
英伟达称，数百款游戏在发布时即可运行，包括《赛博朋克 2077》《堡垒之夜》和《NBA 2K27》。从周五开始，《NBA 2K27》将采用速度更快的 DLSS 5 模型。英伟达表示，该模型可将帧率提升高达 1.4 倍。
Nvidia says hundreds of games run at launch, including Cyberpunk 2077, Fortnite and NBA 2K27. From Friday, NBA 2K27 gets a faster DLSS 5 model. Nvidia says it lifts frame rates by up to 1.4 times.
六家制造商，起售价 2,599 美元：这些笔记本电脑由华硕、戴尔、惠普、联想、微软和微星通过其官网及主要零售商销售。英伟达表示，起售价为 2,599 美元，具体价格因地区而异。
Six makers, from $2,599 The laptops come from ASUS, Dell, HP, Lenovo, Microsoft and MSI, through their websites and major retailers. Prices start at $2,599 and vary by region, Nvidia said.
该系列包括华硕的 ProArt P14 和 P16、戴尔的 XPS 16、惠普的 OmniBook Ultra 和 OmniBook X 14。联想推出 Yoga Pro 9n，微软推出 Surface Laptop Ultra，微星则推出 Prestige N16 和 N16 Flip AI+。未来还将有更多机型推出。
The range includes the ASUS ProArt P14 and P16, the Dell XPS 16 and HP’s OmniBook Ultra and OmniBook X 14. Lenovo has the Yoga Pro 9n, Microsoft the Surface Laptop Ultra, and MSI the Prestige N16 and N16 Flip AI+. More systems will follow.