The Pulse
NVIDIA Expands RTX Spark Software Ahead of Oct. 16 Launch
NVIDIA says RTX Spark laptops will be available October 16, while compact desktops are scheduled to go on sale in November. The company’s October 13 update details new agent features, an open model, a local-compute router and changes to cre
AI.info Team · Written with AI ·
NVIDIA says RTX Spark laptops will be available Friday, October 16, while compact desktops will go on sale in November. The company’s October 13 update names new agent features, an open model, a local-compute router and changes to creative software among the releases tied to the launch.
“The personal computer is the ultimate tool, it’s my ultimate tool, and for a whole generation of people, it’s our ultimate tool,”
said Jensen Huang, NVIDIA founder and CEO.
The announcement describes a broad software push around a platform NVIDIA says combines up to 128GB of unified memory with one petaflop of AI compute. The rollout also reaches Windows on Arm development: NVIDIA says PyTorch with CUDA, native Python and Conda packages are ready, while work continues on the remaining Python packages.
RTX Spark’s agents get local setup and security options
GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi and WorkBuddy are adding or improving RTX Spark support, including simpler local-model setup and options to route work between local and cloud models. NVIDIA says the apps will recommend a model and configuration for the hardware, then guide users through setup. At first boot, a “Getting Started” screen will offer agent apps to install.
The company also points to two security components: Microsoft Execution Containers, which provide operating-system-level controls, and NVIDIA OpenShell, which adds agent setup tools, credential protection and security monitoring. NVIDIA says OpenClaw and GitHub Copilot already support MXC, while Hermes plans to add more granular controls by the end of October. For users who want to mix on-device and cloud inference, NVIDIA’s open-source Switchyard router can direct queries between local and cloud models; Hermes is expected to add it later in October.
Nemotron 3.5 Bolt targets local agent workloads
NVIDIA plans to release Nemotron 3.5 Bolt 74B on October 15, a day before the laptops’ announced launch. The model has 74 billion parameters, with 9 billion active in its mixture-of-experts design, and NVIDIA says it fits into 24GB using 2-bit quantization and quantization-aware training. The company presents it as a model for local agents, balancing speed and capability within that memory limit.
At launch, users will be able to run it through llama.cpp, vLLM, Ollama, Unsloth and LM Studio, or access it through agents including Hermes and OpenClaw. NVIDIA says the model’s training data, weights and recipes will be open source, allowing developers to fine-tune it. The company also says llama.cpp is adding decision-model support and releasing a Windows app for downloading and running models without using the command line.
PAIR and creative apps extend the launch software
NVIDIA PAIR, a free, open-source tool for coordinating AI workloads across compatible computers on a local network, leaves beta on October 13. Its release adds llama.cpp support, direct model downloads from Hugging Face and RTX Spark compatibility. NVIDIA says PAIR runs on Windows, Linux and macOS, letting users distribute workloads across machines rather than relying on a single PC.
Creative software updates span Adobe, NVIDIA Broadcast, ComfyUI and Blender. NVIDIA says Adobe is working toward up to twice the performance in selected tasks across Photoshop, Premiere and Substance 3D, with Photoshop and Premiere optimizations available in beta and Substance 3D updates still to come. Broadcast 2.3 is available to download now and adds video effects, offline processing and controls that let local agents adjust settings and process files through the app.
ComfyUI is adding NVIDIA RTX Video Frame Generation and HDR tools, while Blender’s 5.3 beta adds RTX Spark compatibility. NVIDIA says Blender will add DLSS Ray Reconstruction in November. The company’s list spans software available now, beta features and releases scheduled for later dates, so the October 16 laptop launch does not mean every listed feature arrives on launch day.
Windows on Arm support is a measure of readiness
The developer update addresses a practical hurdle for a Windows-on-Arm PC: getting common programming tools and packages to run natively. NVIDIA says the ecosystem is releasing a PyTorch build with CUDA for Windows on Arm, a native-by-default Python 3.15 build and a native Conda installer with more than 2,000 Conda packages. It also reports that 97% of the top 15,000 Python wheels are ready, while saying it will work with the community on the rest.
That figure puts a boundary around NVIDIA’s claim that RTX Spark is prepared for a broad software lineup. The listed agent, creative and development tools give prospective buyers concrete options, but availability varies by product: some features are downloadable now, others remain in beta, and several are dated for October or November. NVIDIA says the remaining work on Python wheels is still underway.