Microsoft optimizes Windows for Nvidia's Arm-based RTX Spark laptops

At a quieter Nvidia event a day after the media launch, the company detailed the RTX Spark computing platform. According to The Verge, much of the work involved preparing Windows for AI agents and local AI models. Nvidia's selling point is the combination of an Arm CPU, a Blackwell GPU, and unified memory, which is intended to let powerful AI models run on a laptop. The setup experience for an AI agent therefore needs to be quick and easy.

Peter Dawoud, principal program manager at Microsoft, said users opening Windows for the first time on an RTX Spark laptop will see a new Get Started experience. It allows them to find the agent they want, get started with it, install it, and reach it in only two clicks, he said. Agents can be installed and updated through the Windows Store, so users do not have to find updates or determine each agent's dependencies. Dawoud said Microsoft will continue to develop the feature and add more capabilities.

Hermes, Perplexity, and OpenClaw are the first three officially supported agents in Windows at the launch of RTX Spark, The Verge reported. Microsoft is working to add others, which is expected eventually to include Muse, the agent Meta said last week would come to Windows at some point.

In testing, The Verge found that getting started with Hermes was mostly a two-click experience, but OpenClaw did not set up properly at first. The app repeatedly complained that a local gateway had not been installed correctly, and a full uninstall and reinstall was required before everything worked. The Verge said it had used OpenClaw for various tasks over the past week but would need much more testing to understand the value of the local AI support beyond obvious uses such as coding. Large local models remain much slower to respond to prompts than cloud models, so the reporter expected to use the feature more for overnight or background tasks.

Microsoft has also worked behind the scenes to optimize the RTX Spark chips for standard Windows laptop tasks. Unlike Intel or AMD laptops, RTX Spark machines use Microsoft's Prism emulator for applications that do not have a native Arm version. Emulation had been a problem on Windows on Arm for more than a decade, but the launch of Qualcomm's first Snapdragon X series laptops a couple of years ago helped RTX Spark laptops benefit from a wider range of Arm-native apps and Microsoft's improved emulator, according to the report.

Dawoud said emulation is not generic and each platform needs tuning. "Each core is different. How we translate x64 to that core needs to be highly optimized, and for RTX Spark we've done a lot of work to optimize for its microarchitecture," he said. Some of that work has focused on gaming, because CPUs often determine frame pacing and how smooth games feel while GPUs handle rendering. Microsoft ships Prism updates every month to improve compatibility, Dawoud said, and the Windows team continues to work on it.

Memory optimization has been another focus. Microsoft's Surface Laptop Ultra can ship with up to 128GB of unified memory, making it possible to load large models into memory. Dawoud said Microsoft invested in Unified Memory Architecture to let users load as much as possible into the 128GB, while conserving some space so the system remains responsive and applications continue to run well. Microsoft has completed optimizations for how memory works on RTX Spark, including tuning power and performance, maximizing battery life, and ensuring a user can run a game while rendering video in the background. Users can also manually adjust how much unified memory is dedicated to the GPU side with a slider in Nvidia's Windows app. The Verge reported that this slider is useful on RTX Spark laptops with 24GB or 32GB of memory.