Microsoft and NVIDIA Launch RTX Spark Superchip for Next-Generation Windows PCs

Microsoft and NVIDIA unveiled a long-term strategic partnership aimed at creating a new generation of thin-and-light Windows PCs at NVIDIA GTC. Powered by the newly announced NVIDIA RTX Spark superchip, these systems are designed for developers, content creators, and advanced users, with a strong emphasis on running AI agents and AI-powered workflows locally.

The collaboration combines Microsoft's Windows platform with NVIDIA's latest computing and graphics technologies, enabling PCs to evolve from traditional productivity tools into intelligent, AI-assisted computing companions.

RTX Spark Platform and Windows Integration

The NVIDIA RTX Spark superchip features up to 6,144 CUDA cores based on the Blackwell architecture, fifth-generation Tensor Cores with FP4 precision support, and a 20-core Arm-based CPU developed in partnership with MediaTek. Combined with up to 128GB of unified memory and up to 1 petaflop of AI computing performance, the platform is supported by a range of Windows 11 optimizations designed specifically for its heterogeneous architecture.

Performance and Power Optimization

Microsoft has introduced several platform-level enhancements to improve efficiency, responsiveness, and battery life on RTX Spark-powered devices.

• Workload Profile Scheduling (WPS): Optimized for RTX Spark, WPS enables Windows to intelligently distribute workloads across all 20 CPU cores, improving efficiency for tasks ranging from everyday productivity to software development and AI processing.
• Microsoft Power and Thermal Framework (MPTF): Integrated directly into the platform, MPTF provides consistent thermal and power management across different device designs, helping maintain sustained performance while preserving battery endurance.
• Graphics and AI APIs: The platform leverages DirectX 12 enhancements for neural rendering and advanced ray tracing. Windows ML has also been optimized to provide direct access to NVIDIA TensorRT acceleration for AI applications.

Enhanced Unified Memory Architecture

To take full advantage of the platform's unified memory design, Microsoft has updated Windows memory management to improve how CPU and GPU resources share memory.

• Memory Capacity: Supports up to 128GB of unified memory.
• Expanded GPU Access: Windows now allows a larger portion of system memory to be dynamically allocated to the GPU when required.
• AI Workloads: Enables execution of local large language models containing up to 120 billion parameters with context windows reaching 1 million tokens.
• 3D Content Creation: Supports rendering and manipulation of complex 3D scenes exceeding 90GB in size.
• Adaptive Page Management: Shared memory regions can dynamically adjust page sizes to improve workload performance while maintaining efficient CPU-GPU data exchange.

Improved Prism Emulation

Microsoft has also enhanced its Prism emulation technology to improve compatibility and performance for legacy x86 applications running on Arm-based RTX Spark systems.

Prism continues to support both 32-bit and 64-bit x86 software and now includes further optimizations tailored specifically for the RTX Spark architecture. Building on earlier support for AVX and AVX2 instruction sets, these improvements help reduce performance overhead for non-native applications, including development tools, creative software, and PC games.

Secure Foundation for Local AI Agents

To address privacy and security concerns associated with cloud-based AI services, Microsoft and NVIDIA used Microsoft Build to demonstrate a new platform for running autonomous AI agents directly on Windows devices.

The platform is built on native Windows security technologies that provide operating system-level identity management, containment, policy enforcement, and administrative controls. On top of this foundation sits the NVIDIA OpenShell runtime, which gives users granular control over the data accessible to AI agents, routes requests according to compliance requirements, and masks sensitive personal information before interacting with external cloud-based models.

Several open-source projects, including Hermes Agent and OpenClaw, are already incorporating these Windows security capabilities. This enables AI agents to perform tasks such as cross-application workflows, code generation, and semantic file searches entirely on-device. Microsoft also plans to integrate these AI agent experiences directly into the Windows taskbar in future updates.

Expanding Software and Gaming Support

RTX Spark-powered systems launch with a growing ecosystem of applications optimized for Arm-based Windows devices, spanning creative workloads, software development, AI research, and gaming.

Creative Applications

Adobe has redesigned several of its flagship applications to run natively on the platform, including Photoshop, Premiere Pro, and Substance 3D Painter and Stager. A new video processing pipeline combines unified memory and TensorRT acceleration to improve AI-powered features such as Firefly Generative Fill and Generative Extend, delivering up to twice the performance.

Additional native creative applications include Blender 5.3 with DLSS 4.5 Ray Reconstruction support, DaVinci Resolve, Maxon Cinema 4D, Topaz Photo AI, CapCut, and the Affinity suite by Canva.

Development and AI Tools

The platform also expands support for technical and AI-focused workloads. MATLAB now officially supports Windows on Arm, while AI developers gain access to CUDA-accelerated versions of PyTorch, llama.cpp, Hugging Face frameworks, Unsloth, Kohya, and node-based tools such as ComfyUI, which supports RTX Video and 4x Frame Generation technologies.

Gaming Ecosystem

Gaming support is strengthened through native compatibility with Epic Games' Easy Anti-Cheat and BattlEye technologies, as well as integration with the Xbox PC app. These improvements help x86 titles run efficiently through the enhanced Prism emulation layer.

Several major games are also being optimized for the platform, including League of Legends, VALORANT, PUBG: Battlegrounds, Alan Wake 2, Naraka: Bladepoint, Pragmata, and War Thunder.

Hardware Partners and Device Lineup

As part of the Copilot+ PC ecosystem, RTX Spark devices combine a dedicated NPU for everyday AI tasks with NVIDIA's Blackwell GPU for more demanding workloads. Microsoft's Surface Laptop Ultra will be among the first devices powered by the new platform.

Multiple hardware manufacturers are preparing laptops and compact desktop systems scheduled to launch this fall.

Jensen Huang, Founder and CEO of NVIDIA, said:

“The PC is undergoing a major transformation. For decades, users interacted with computers by launching applications, clicking through menus, and entering commands. With RTX Spark and Microsoft Windows, computing becomes more conversational—users can simply make a request, and the system carries out the task.

RTX Spark brings together NVIDIA’s key technologies, including CUDA, RTX graphics, and its AI software stack, within a single superchip. It enables local AI agents, advanced AI models, creative applications, and RTX-powered gaming experiences to run directly on a laptop. This represents a new era of personal computing—the personal AI computer.”