What’s Changing Ahead of the RTX Spark Launch? A Guide to Supported Apps, from Local AI to Gaming

Start up an RTX Spark laptop, choose an AI agent on the first screen, and install it. Supported apps guide you through an RTX Spark configuration without making you hunt down models and settings yourself. NVIDIA’s announced ecosystem updates cover everything from that initial setup to video editing, gaming, and development.

NVIDIA and Microsoft announced that RTX Spark laptops will launch on October 16, 2026. Software and model providers are preparing support to coincide with the release. This article organizes the announcements by use case. Peak performance and compatibility vary by app version, settings, and model, so check each product’s guidance before using it.

Making Local AI Agents Easier to Get Started With#

RTX Spark is a PC advertised with up to 128GB of unified memory and up to 1 petaflop of AI compute performance. NVIDIA has outlined updates to simplify installation for agent apps including GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi, and WorkBuddy. Supported apps will recommend local models and settings for RTX Spark, reportedly allowing setup in just a few steps.

For security, NVIDIA OpenShell works alongside Microsoft Execution Containers (MXC). MXC provides OS-level isolation, while OpenShell simplifies deployment, protects credentials, and monitors security. OpenClaw and GitHub Copilot already support MXC. Hermes Agent is also scheduled to integrate granular OpenShell controls and MXC by the end of October.

For tasks that local models cannot handle alone, there is also the option to use cloud models alongside them. NVIDIA’s open-source model router, Switchyard, routes requests between local and cloud models. Hermes support is planned for the second half of October, while GitHub Copilot and other apps incorporate their own routing features.

One model introduced for local use is Nemotron 3.5 Bolt 74B-A9B. This mixture-of-experts (MoE) model has 74 billion total parameters and 9 billion active parameters, with a design that uses 2-bit quantization to fit within 24GB. It can reportedly run through llama.cpp, vLLM, Ollama, Unsloth, and LM Studio, and can also be used through Hermes and OpenClaw. NVIDIA says it will release the model as open source, including its training data, weights, and recipes.

App-Specific Optimizations for Creative Work#

Adobe is optimizing Photoshop, Premiere, Substance 3D, and other apps for RTX Spark, claiming speed improvements of up to 2× for some creative tasks. These cover GPU-accelerated compositing in Photoshop, as well as playback, rendering, color grading, and AI features in Premiere. In Substance 3D Painter, the updates are intended to improve high-resolution mesh baking and loading, saving, and exporting large projects.

Availability varies by feature. AI feature optimizations for Photoshop and Premiere are reportedly available in beta, while the Substance 3D updates are coming soon. The figure of up to 2× comes from NVIDIA and Adobe’s announcements, so you will need to test your own assets and workflows to see whether your projects achieve similar gains.

NVIDIA Broadcast 2.3 is available to download and runs natively on RTX Spark. Alongside new Video Super Resolution, Video Frame Generation, and studio audio profiles, it supports offline processing to apply effects to prerecorded files. An MCP server and skills have also been added, allowing AI agents to control effects and settings and process files.

ComfyUI is getting RTX Video Frame Generation and RTX Video HDR nodes. Frame generation increases video frame rates on supported GPUs, while the HDR feature converts SDR footage to HDR. Combined with RTX Video Super Resolution, these features let you handle everything from upscaling to finishing within the same workflow. NVIDIA’s example shows a 1080p, 30fps SDR clip created with LTX 2.5 being converted into 60fps HDR video, then exported as EXR for color adjustment.

Blender supports RTX Spark in the version 5.3 beta. DLSS Ray Reconstruction is scheduled to arrive in November, with the aim of making viewport interactions and previews smoother. Before adopting it, check compatibility with your current plug-ins and production environment.

Development Tools Expand to Windows on Arm#

For developers, support is progressing for PyTorch, Python, CUDA, and CUDA libraries. NVIDIA and Microsoft announced a PyTorch-CUDA build for Windows on Arm, a native build of Python 3.15, and a native installer that includes more than 2,000 Conda packages. They also say that 97% of 15,000 major Python wheel packages are now supported.

Windows Developer Config will also be available to set up a development environment on Windows in one go. With Windows Subsystem for Linux, developers can work locally with the packages they use for cloud deployment. To supplement agents’ CUDA knowledge, NVIDIA is also releasing open-source “NVIDIA skills” that let agents use CUDA-X tools and libraries.

Among coding agents, GitHub Copilot supports MXC and is updating its HydraFusion model router to handle local models. Microsoft is making coding models such as MAI 1.1 Coding Flash available locally and adding llama.cpp as a Windows ML backend. Using local models can reduce cloud requests, but model availability and response quality depend on the configuration.

llama.cpp is also offering a Windows GUI app called Llama Windows. It launches from the Windows 11 system tray and lets users download and run models without using the command line. Supported models can be selected from Hugging Face Hub, so a sensible starting point is to try the setup process with a small model, confirm that it works, and then choose one suited to your needs.

Support Announced for Games and AI Characters#

NVIDIA says hundreds of popular games will be available at launch. The announcement lists titles including *Cyberpunk 2077*, *Fortnite*, *Genshin Impact*, *Hogwarts Legacy*, *Honkai: Star Rail*, *Marvel Rivals*, and *NBA 2K27*. Alongside DLSS, Reflex, ray tracing, and G-SYNC, the company says it has also worked on anti-cheat support for multiplayer games. Check title-specific information after launch for details on supported features and performance.

DLSS 5 was announced for release on RTX Spark’s launch day. Its new AI model is said to be twice as fast as the previous model and will debut in *NBA 2K27* on October 16. NVIDIA says it can increase frame rates in that title by up to 1.4×. Support is scheduled for *Hogwarts Legacy* on November 5 and *Where Winds Meet* in late November. An update adding DLSS 5 support to RTX Remix is planned for October 23.

For in-game AI characters, Nemotron 3.5 Bolt E4B 4GB is scheduled for release on October 23. This small language model is designed for voice conversations, multi-turn tool use, and responses informed by an app’s state. In SkyrimNET, it can power conversations with NPCs. SkyrimNET supports both cloud and local inference, so you will need to test local operation in your own environment to assess usability and system load.

Start by Checking Support for the Apps You Plan to Use#

To try these announcements, first choose a use case and check the relevant apps. For AI agents, look into the apps available for setup, local models, and MXC support. For video production, candidates include Broadcast 2.3 and ComfyUI’s new nodes; for development, Windows Developer Config and Llama Windows are options. For gaming, check which RTX features and anti-cheat systems are supported by the titles you want to play.

Some release dates are provisional, and availability may vary by region and app update status. Compare NVIDIA’s announcement with each software provider’s release notes, and start with the features relevant to your work. If you are considering RTX Spark, check the requirements for the apps you want to use after launch and list the models and software you will need in advance to make setup smoother.

Source: https://blogs.nvidia.co.jp/blog/local-ai-rtx-spark-laptops-software-updates/