What RTX Spark Is and Why Its ARM SoC Matters
RTX Spark is NVIDIA’s new ARM-based PC platform that combines a 20-core Grace CPU, a Blackwell RTX GPU, and unified LPDDR5X memory to deliver high-frame-rate gaming and AI performance in ultrathin Windows laptops and compact desktops. Designed with MediaTek, the RTX Spark architecture targets Windows ARM laptops with a single piece of silicon manufactured on TSMC’s 3 nm process and containing 70 billion transistors. NVIDIA says RTX Spark systems can be as thin as 14 mm, yet still run mainstream games at around 100 frames per second at 1440p resolution. Unlike earlier Windows on ARM efforts, Spark leans on the full RTX software stack, including CUDA, TensorRT, DLSS, Reflex, and ray tracing, to make ARM SoC gaming feel closer to existing x86 laptops while promising better efficiency and battery life.
Grace CPU Cores: Ten Cortex-X925 and Ten Cortex-A725
At the heart of RTX Spark sits a custom Grace CPU configuration that mirrors high-end smartphone SoCs: ten Cortex-X925 performance cores paired with ten Cortex-A725 efficiency cores. This 10+10 layout gives the platform 20 Grace CPU cores that can swing between high single-thread speed and low-power background processing, similar to how modern phones juggle intensive apps and standby tasks. According to Gizmochina, this design reflects NVIDIA "taking ideas that worked well in flagship phones and scaling them up for PCs." The performance-focused Cortex-X925 cluster handles gaming, content creation, and heavy AI inference, while the Cortex-A725 cluster takes care of light workloads to extend battery life. This smartphone-style core strategy is central to RTX Spark architecture, allowing Windows ARM laptops to balance responsiveness with all-day mobility more effectively than past PC-only core designs.
Blackwell GPU Performance and Unified Memory for AAA Gaming
The RTX Spark platform pairs its Grace CPU cores with a Blackwell GPU featuring 6,144 CUDA cores and fifth-generation Tensor Cores, claimed to reach about 1 PFLOP of AI performance using FP4. NVIDIA estimates Blackwell GPU performance in Spark to be similar to a GeForce RTX 5070 Mobile, enabling ray tracing, DLSS, and high refresh-rate gaming in thin designs. The GPU and CPU share up to 128 GB of LPDDR5X unified memory, delivering about 300 GB/s of bandwidth over an NVLink C2C interconnect instead of separate VRAM and system RAM pools. This unified design is especially important for AI and content workloads, with NVIDIA highlighting support for 3D scenes up to 90 GB, 12K video editing, and AI models with 120 billion parameters and a one-million-token context window running locally on ARM SoC gaming laptops.
Smartphone-Style Design Meets Windows ARM Laptops
RTX Spark borrows smartphone chipset design patterns and applies them to Windows ARM laptops and mini PCs: a mixed-core Grace CPU, a powerful integrated GPU, and unified memory. The 10 Cortex-X925 and 10 Cortex-A725 core split resembles flagship mobile SoCs, while the Blackwell GPU brings desktop-class RTX features, including ray tracing and DLSS, into ultrathin PC form factors. NVIDIA and Microsoft have worked together to optimize Windows for Spark, especially for agentic and generative AI tasks that benefit from the platform’s PFLOP-scale AI throughput. Hardware partners such as ASUS, Dell, HP, Lenovo, MSI, Acer, and Microsoft’s Surface division plan to release commercial laptops around autumn, many aiming for 14 mm designs with Tandem OLED G-SYNC displays. With this move, NVIDIA is no longer just a GPU supplier; it is entering the full Windows on ARM PC market with a tightly integrated SoC and software stack.
