Nvidia Redefines PCs with the Ultra‑Efficient RTX Spark CPU

At the GTC Taipei 2026 event, Jensen Huang announced a new class of Windows PCs built around Nvidia’s RTX Spark super‑chip, a collaboration between Nvidia and Microsoft that aims to redefine the hardware foundation of personal computers.

The RTX Spark chip reuses the GB10 silicon from Nvidia’s DGX Spark and is based on an ARM architecture; the CPU portion is designed by MediaTek. It integrates a 20‑core Nvidia Grace CPU, a Blackwell RTX GPU with 6144 CUDA cores, fifth‑generation Tensor Cores supporting FP4 precision, and NVLink‑C2C interconnects.

Key specifications include up to 1 petaflop of AI compute, 128 GB of unified memory, and a design that supports both ultra‑thin laptops (as thin as 14 mm, weighing 3 lb, with 14‑16 inch OLED displays and G‑SYNC) and compact high‑efficiency desktops targeting AI agents, creative workloads, and gaming.

The platform is positioned as an "AI user experience": users interact with the PC via natural language rather than traditional UI, leveraging local AI agents for tasks such as image/video generation, semantic file search, and cross‑application workflows. Nvidia’s OpenShell runtime and new Windows security primitives provide a secure, privacy‑preserving environment for running agents.

Performance claims highlight that generative AI tools in Photoshop and Premiere can achieve up to 2× speed‑up thanks to the fifth‑generation Tensor Cores, while the unified memory architecture eliminates the bottleneck of discrete GPU memory for high‑resolution video and complex 3D geometry.

For gamers, the ARM‑based CPU is paired with Microsoft’s Prism emulator and collaborations with anti‑cheat providers (Easy Anti‑Cheat, BattlEye) to ensure compatibility with popular titles such as League of Legends, VALORANT, PUBG, and AAA games like Alan Wake 2, supporting DirectX 12 neural rendering and ray tracing.

Overall, Nvidia presents RTX Spark as a versatile, high‑performance, and energy‑efficient platform that bridges the gap between traditional PCs and dedicated AI workstations, enabling offline execution of large‑parameter models and reducing reliance on cloud resources.