Microsoft and NVIDIA announced NVIDIA RTX Spark at NVIDIA GTC, unveiling what both companies describe as the most powerful and efficient thin-and-light Windows PCs ever built. The chip delivers 1 petaflop of AI performance, up to 6,144 Blackwell RTX cores, 20 power-efficient Arm-architecture CPU cores, and up to 128GB of unified memory, combining GPU, NPU, and CPU into a single silicon package purpose-built for AI-native computing.
RTX Spark targets developers, creators, and power users who need to run advanced AI workloads, graphically intensive creative applications, and agent-based tasks locally, without relying on cloud infrastructure. Microsoft worked directly with NVIDIA to optimize Windows scheduling, memory management, and thermal controls for the chip’s heterogeneous architecture. Workload profile scheduling enables the Windows scheduler to distribute tasks more efficiently across all 20 cores, while the Microsoft Power and Thermal Framework maximizes performance-per-watt under sustained workloads.
The unified memory architecture unlocks a significant capability jump for local AI. Systems with up to 128GB allow developers to load large language models entirely on-device, run complex rendering projects, and build agentic workflows that previously required cloud compute. Microsoft separately confirmed that NVIDIA will bring NVIDIA OpenShell to Windows, with Hermes Agent and OpenClaw integrating new Windows security and containment primitives for safe, local agent execution.
App ecosystem support at launch includes Blender, DaVinci Resolve, Photoshop, Premiere, CapCut, GitHub Copilot, Claude Code, Cursor, ComfyUI, and CUDA-accelerated PyTorch. Riot Games confirmed League of Legends and VALORANT will run on RTX Spark, and PUBG: Battlegrounds, Alan Wake 2, and Naraka: Bladepoint also confirmed compatibility.
RTX Spark-powered Copilot+ PCs will launch this fall from Microsoft Surface, ASUS, Dell, HP, Lenovo, and MSI. Microsoft also announced a longer-term roadmap scaling Windows from RTX Spark laptops up to the NVIDIA DGX Station for Windows, powered by the GB300 Grace Blackwell Ultra Desktop Superchip, which the company frames as putting a trillion-parameter AI supercomputer on every enterprise desk.
