NVIDIA 为 RTX Spark 笔记本铺开本地 AI 软件生态

NVIDIA RTX Spark is a new beginning for the PC — a reinvention combining the capabilities of a gaming PC, a creative studio and a personal AI supercomputer into a single superchip. With up to 128GB of unified memory and 1 petaflop of AI compute, RTX Spark delivers the full stack of NVIDIA RTX and CUDA technologies for creators, developers and gamers, while unlocking a new paradigm in personal computing via powerful local AI agents.

Last week, NVIDIA and Microsoft announced that RTX Spark laptops will be available on Friday, Oct. 16. Ahead of the launch, NVIDIA and its ecosystem partners are rolling out new software features and performance optimizations to expand what’s possible with local AI on RTX Spark. 

New and upcoming RTX Spark updates — among the thousands of apps, games and models available at launch — include: 

• New agent features, models and tools that enhance local agents on RTX Spark: 
  • Updates to top agent apps including GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi and WorkBuddy streamline setup on RTX Spark, improve security with OpenShell and Microsoft Execution Containers (MXC), and enable hybrid local and cloud AI routing.
  • The NVIDIA Nemotron 3.5 Bolt 74B model, purpose-built for agentic AI on RTX Spark, releases on Thursday, Oct. 15. 
  • NVIDIA PAIR moves out of beta and adds llama.cpp support.
• Top creative apps optimized for RTX Spark:
  • Up to 2x performance is coming soon to Adobe’s top apps. NVIDIA and Adobe have collaborated to rearchitect Photoshop for GPU-accelerated compositing and Premiere for playback, rendering, effects and more.
  • The NVIDIA Broadcast 2.3 release brings new Video Super Resolution, Video Frame Generation, offline file processing and agent Model Context Protocol (MCP) controls to the app.
  • NVIDIA RTX Video Frame Generation and high dynamic range (HDR) capabilities are available now in ComfyUI to accelerate AI generative video workflows.
  • NVIDIA DLSS Ray Reconstruction will be available in Blender 5.3 in November to improve interactivity in the viewport.
• Developer tools are now ready for RTX Spark:
  • Top developer frameworks are now available for RTX Spark, including PyTorch, Python and CUDA, as well as all CUDA libraries. 
  • GitHub Copilot, Pi and other coding agents are now optimized for RTX Spark.
  • Llama.cpp adds support for decision models, and releases a Windows app.
• Hundreds of top games and anti-cheat are supported on RTX Spark, plus:
  • A 2x faster DLSS 5 AI model arrives Friday, Oct. 16, and a new Performance Mode for DLSS Frame Generation is available today via the NVIDIA App beta.
  • Nemotron 3.5 Bolt E4B 4GB, the third-generation small language model (SLM) for NVIDIA ACE AI game characters, releases Friday, Oct. 23.
  • The latest RTX Remix update, including DLSS 5 support, releases on Friday, Oct. 23.

Advancing Local AI Agents With New Apps, Models and Tools

AI is changing the way people interact with PCs. Agent popularity skyrocketed this year, with projects such as OpenClaw and Hermes topping popularity charts on GitHub. Local large language models (LLMs) surpassed the intelligence threshold to power agents, with models like Qwen 3.8-Flash-Next catching up to frontier cloud models released just a few months ago. As a result, local LLM downloads are already 5x higher than all of last year.

From models to agent harnesses to security to tools, NVIDIA is partnering closely with the local AI ecosystem to make RTX Spark the ideal hardware to run local agents. 

Local AI Agents Made Easy On RTX Spark

From the first boot of RTX Spark, users will be greeted with a “Getting Started” screen to choose and install their agent app.

These agents will come with new, secure sandboxing technology thanks to NVIDIA OpenShell and MXC. MXC delivers operating system-level security, while OpenShell makes it simpler for agents to adopt MXC and adds advanced capabilities such as credential protection and security monitoring. Hermes Agent will integrate granular OpenShell controls and MXC by the end of the month. OpenClaw and GitHub Copilot now support MXC.

After installing an agent, setting up local AI is also now seamless in GitHub Copilot, Hermes Agent, OpenClaw, Perplexity, Pi and WorkBuddy. The apps will recommend an optimized local model and settings configuration for RTX Spark, and set it all up for users in a few clicks.

Many users want the best of both worlds — local models for free iteration, and cloud models for frontier intelligence. NVIDIA Switchyard, an open source model router library, enables agents to intelligently route queries between local and cloud LLMs, allowing the user to minimize cloud usage but still obtain similar results to just running on cloud. Hermes will be adding Switchyard later in October, and other agents like Perplexity Computer and GitHub Copilot have integrated their own routing solutions.

Software providers have made it simple for agents to connect to top Windows apps, with easy-to-use MCP libraries and command line interfaces (CLIs). NVIDIA has also now made available MCPs for its applications, including NVIDIA App, Broadcast, PAIR, Omniverse, RTX Remix and a large library of CUDA-X Skills. 

NVIDIA is also releasing AI playbooks for RTX Spark, available on Friday, Oct 16. Users can set up their first agent, build an AI chief of staff, configure their system for development or design a creative assistant.

Meet Nemotron 3.5 Bolt 74B: Built for Local Agents on RTX Spark

NVIDIA Nemotron 3.5 Bolt 74B-A9B 24GB, available Thursday, Oct. 15, is a new LLM purpose-built for running local agents on RTX Spark. 

This model is a 74B parameter, 9B-active-parameter mixture-of-experts (MoE) model, packed into 24GB using 2-bit quantization with advanced quantization-aware training techniques to retain quality. It aims to achieve the optimal balance between speed and intelligence, being faster than most popular dense models while smarter than most popular MoE models with a similar memory footprint.

Users will be able to run Nemotron 3.5 Bolt 74B at launch through llama.cpp, vLLM, Ollama, Unsloth and LM Studio. It will also be directly accessible via agents such as Hermes and OpenClaw.  

Like other Nemotron models, Bolt 74B is fully open source — including training data, weights and recipes — so developers can fine-tune and customize their own agents.

The Perfect PAIR

Local AI doesn’t have to run on just one PC. NVIDIA PAIR, a personal AI router, is officially moving out of beta today, making it easier to tap into the total AI compute of compatible PCs on a local network. The free, open source software tool automatically discovers and connects devices, installs AI models and inference engines, and distributes AI workloads across available systems.

The official release adds llama.cpp support, direct model downloads from Hugging Face, customizable inference engine settings and RTX Spark support, along with a variety of inference and user experience improvements. 

PAIR works on Windows, Linux and macOS. Developers can contribute to the open source project via Github — just one month after the initial PAIR release, the community has already contributed over 10 features and fixes to the release. 

Sparking Creativity

RTX Spark brings over 130 accelerated apps, CUDA for fast video and image generation, and RTX technologies such as DLSS, NVENC and OptiX to advanced content creation workflows — all in a thin and light laptop with on-the-go performance.

Adobe Accelerates Creative Applications for RTX Spark

Adobe is optimizing its industry-leading creative applications, including Photoshop, Premiere and Substance 3D, for RTX Spark, delivering up to 2x faster AI, editing, color and effects across creative workflows. 

In Photoshop, GPU-accelerated compositing makes complex image editing faster and more responsive, with improvements to the Remove Tool and effects such as Light Leaks. Premiere accelerates demanding video editing tasks, including playback, rendering, color grading and AI-powered features such as Enhance Speech and Object Mask. Meanwhile, Substance 3D Painter speeds up texture creation workflows, from baking high-resolution mesh maps to opening, saving and exporting large projects.

The updates will be rolling out in phases, with AI feature optimizations in Photoshop and Premiere now ready in beta, and with Adobe Substance 3D updates coming soon. Opt in to the beta to get access to the RTX Spark-optimized builds. 

Latest NVIDIA Broadcast Brings New AI Effects to Live and Recorded Content

Available for download now, NVIDIA Broadcast 2.3 runs natively on RTX Spark, and adds new RTX Video Super Resolution and Frame Generation AI effects and Studio Voice profiles.

A new Offline File Processor lets creators apply Broadcast effects to prerecorded content, with many effects running at higher quality since there isn’t a need to maintain real-time performance. The update also introduces a new MCP server and skill that lets local AI agents adjust effects, change settings and process files directly through the Broadcast app.

RTX Video Brings Frame Generation and HDR to ComfyUI

ComfyUI is expanding its RTX Video tools with two new nodes: RTX Video Frame Generation and RTX Video HDR. It’s also providing the ability to export EXR files, giving creators more ways to enhance AI-generated video.

RTX Video Frame Generation increases video frame rates for smoother motion for GeForce RTX 40 Series GPUs and newer, while RTX Video HDR converts standard dynamic range (SDR) footage to HDR for greater flexibility in color grading. Combined with RTX Video Super Resolution, these tools enable creators to upscale, enhance and refine video within a single ComfyUI workflow.

For example, creators can combine LTX 2.5 with RTX Video Frame Generation and HDR to transform a 1080p, 30 frames-per-second (fps) SDR clip into a smoother 60 fps HDR video exported as EXR files ready for color grading. 

Blender Adds DLSS 4.5 Ray Reconstruction

Blender has released a beta version of its popular 3D editor, adding compatibility for RTX Spark. Download Blender version 5.3 to get access.

Coming next month, Blender will also be adding support for DLSS Ray Reconstruction, allowing for real-time updates in the viewport so artists can navigate a scene while previewing it in real time. 

Code, Text and Deploy From RTX Spark

Developers are moving from writing code to directing AI agents. This new development paradigm is requiring many to spread their work across their laptop, a cloud LLM and a cloud virtual machine, where they test and prototype. 

RTX Spark enables developers to work on one machine, running all their workloads and libraries, testing and prototyping, and deploying. They can now do so directly in Windows with the new Windows Developer Config, which enhances the user experience for devs and installs all the top developer tools — or via Windows Subsystem for Linux, where they can get access to the same packages they’ll use to publish to the cloud.

Developer Tools Are RTX-Spark Ready

NVIDIA has partnered with Microsoft to ensure that developer tools are ready for RTX Spark. Alongside RTX Spark, the community is releasing:

• The first PyTorch-CUDA build for Windows on Arm.
• The first Python build that is native by default on Windows on Arm, with Python 3.15.
• The first native Conda installer for Windows on Arm, with over 2,000 Conda packages.
• 97% of the top 15,000 Python wheels are ready, thanks to work with the ecosystem. NVIDIA will be working closely with the community to ready the remaining wheels.

Development is now mostly done by agents. NVIDIA Skills — an open source GitHub repository that teaches agents how to use NVIDIA CUDA-X tools and libraries — lets users develop apps that are accelerated on NVIDIA technologies, even without knowing anything about CUDA.

Coding Agents Are Ready for Local AI

In addition to GitHub Copilot — one of the most popular coding agents, used by millions worldwide — adding support for MXC, it’s updating its model router HydraFusion with support for local models. This enables developers to use their local hardware to reduce token consumption while tapping into frontier intelligence.

Microsoft is making coding models, such as MAI 1.1 Coding Flash, available locally, and adding llama.cpp as a backend to Windows ML — extending the supported models and enabling the latest NVIDIA performance optimizations.

Get started with GitHub Copilot running locally when RTX Spark launches.

Llama.cpp Adds Support for Decision Models, Releases a Windows App

Decision models are the latest breakthrough in AI development. They give AI agents a near-instant way to make decisions, without lengthy reasoning loops. They can choose what to do next, from classifying requests to selecting the next action, up to 200x faster than traditional agents. Decision models are now supported in llama.cpp and Unsloth on RTX Spark and GeForce RTX GPUs.

Llama.cpp is also releasing today Llama Windows, a Windows-native companion app for llama.cpp, enabling an easy-to-use user interface to download and run models locally, without having to go into the command line. This app, like llama.cpp, is optimized for RTX Spark. Learn more on the Hugging Face blog.

Game On: RTX Spark Launches With Hundreds of Top Games and DLSS 5

RTX Spark is ready for gaming, with hundreds of top titles available when laptops launch on Friday, Oct. 16, including CONTROL Resonant, Cyberpunk 2077, Fortnite, Gears of War: E-Day, Genshin Impact, Hogwarts Legacy, Honkai: Star Rail, Marvel Rivals, NARAKA: BLADEPOINT, NBA 2K27, World of Warcraft and many more.

NVIDIA has worked to bring the full stack of RTX technologies, including DLSS, Reflex, Remix, G-SYNC and ray tracing, to the new platform, along with collaborating with game developers to enable anti-cheat support for popular multiplayer games. Learn more about gaming on RTX Spark. 

DLSS 5 Brings Neural Rendering to RTX Spark

NVIDIA DLSS 5 with 3D-Guided Neural Rendering is also coming to RTX Spark on launch day. The technology uses AI to enhance lighting and materials with lifelike detail, making game worlds and characters more realistic while preserving developers’ artistic vision. A new, 2x faster DLSS 5 AI model debuts in NBA 2K27 on Friday, Oct. 16, boosting frame rates by up to 1.4x. The faster model will also power DLSS 5 in Hogwarts Legacy starting Thursday, Nov. 5, and Where Winds Meet in late November.

DLSS 5 is also coming to NVIDIA RTX Remix on Friday, Oct. 23, so modders can add better materials, lighting and shadows to classic characters and environments, as shown from modder Sparkles/Kim2091 in her in-development project, Fallout 4 RTX.

Read more about DLSS 5 on RTX Spark on GeForce.com.

Game Agents Running Locally on RTX Spark

Nemotron 3.5 Bolt E4B 4GB is the third-generation SLM for NVIDIA ACE, the suite of technologies for building in-game AI agents that deliver next-generation gaming experiences. The first-generation model was in a public beta earlier this year for PUBG, allowing players to team up with AI teammates, while the more efficient second-generation model previewed in Aniimo at Gamescom, turning standard creatures into voice-responsive companions.

Nemotron 3.5 Bolt E4B combines natural conversation with multi-turn tool use, instruction following and grounding in application state to power fast local voice and device assistants, interactive characters and embedded interfaces.

Designed for a low memory footprint and fast time to first token, this model enables responsive local experiences while leaving headroom for applications and other AI models. Its new conversational mode helps keep responses concise and speech-ready, and developers can turn reasoning on or off as needed. 

Starting Friday, Oct. 23, test drive the third-generation Nemotron 3.5 Bolt E4B 4GB inside SkyrimNET, an AI nonplayable character (NPC) modding platform by MinLL for The Elder Scrolls V: Skyrim. SkyrimNET transforms every NPC into an unscripted, interactive character equipped with personal memory, voice and autonomy for dynamic conversations and context-aware quests generated on the fly.

Now compatible with RTX Spark, SkyrimNET supports both cloud and local inference across multiple models, with the latest Nemotron models delivering best-in-class performance for efficient and fast in-game interactions. Download the build on GitHub.

Accelerating Modding Workflows in RTX Remix

As AI agents continue to help modders make their favorite classic games compatible with the RTX Remix renderer, NVIDIA is streamlining asset updates with a major overhaul to its AI-powered texture generation workflow. Launching Friday, Oct. 23, as part of Remix 1.6, ComfyUI Version 2 replaces complex REST application programming interface (API) node connections with a dedicated AI tools layout directly in the sidebar.

This update allows modders and artists to generate physically based rendering materials natively inside RTX Remix with zero API setup required. A new batch processing feature makes it easy to scale beyond single-texture experimentation to full modding projects, while an intuitive presets system lets non-technical creators adjust key settings without touching the node graph. 

Subscribe to the NVIDIA Local AI newsletter. Follow NVIDIA RTX Spark on X, Instagram, TikTok and Facebook.

See notice regarding software product information.