What RTX Spark Is and Why It Matters
NVIDIA RTX Spark is an Arm-based system-on-chip for Windows PCs that combines 20 custom Grace CPU cores, a Blackwell RTX GPU, and up to 128GB of unified LPDDR5X memory into a single platform aimed at premium laptops and compact desktops, bringing a smartphone-style hybrid architecture to mainstream computing. Announced by Jensen Huang during the Computex keynote, RTX Spark is NVIDIA’s second attempt at Windows-on-Arm, but this time it arrives with full RTX software support and hardware derived from the proven GB10 data center SoC. Microsoft is treating it as a flagship "new era of PC" platform for Windows, and major OEMs are lining up 14‑ to 16‑inch aluminum ARM Windows laptop designs around it. Together, these moves position RTX Spark ARM processors as direct rivals to x86 chips from Intel and AMD in mobile and small form factor systems.
Inside the 20-Core Grace CPU: Smartphone Logic for PCs
At the heart of RTX Spark is a 20-core Grace CPU designed with MediaTek, split into ten Arm Cortex‑X925 performance cores and ten Cortex‑A725 efficiency cores. This mirrors the hybrid big.LITTLE style seen in smartphone chipsets, but scaled for PC-class sustained workloads and multi-tasking. High-intensity tasks like game engines, compilers, and AI inference threads can sit on the Cortex‑X925 cores, while background services and light apps are pushed to the efficiency cluster to save power. According to Gizmochina, this Arm layout is NVIDIA "taking ideas that worked well in flagship phones and scaling them up for PCs." If Windows and application schedulers use these NVIDIA Grace CPU cores well, RTX Spark could deliver strong burst performance while keeping thermals and battery drain in check, a pattern that has powered mobile devices for years but has been slower to reach traditional laptops.
Blackwell GPU Integration and Unified Memory for AI
RTX Spark’s other defining feature is Blackwell GPU integration on the same SoC as the Grace CPU. The chip includes a Blackwell-based accelerator with 48 streaming multiprocessors and 6,144 CUDA cores, delivering around 1 PFLOP of FP4 AI performance, a figure comparable to a GeForce RTX 5070-class GPU in a thin laptop envelope. Instead of discrete VRAM, up to 128GB of LPDDR5X memory is shared between CPU and GPU as a unified pool. That means large AI models and complex graphics scenes do not have to shuttle data across a narrow PCIe link or duplicate buffers between system and graphics memory. For AI workloads, unified memory plus Blackwell GPU integration could cut latency and reduce overhead, while also saving power compared with separate CPU and GPU packages, particularly in always-plugged SFF desktops and creator laptops running local generative models.
Target Devices: Premium ARM Windows Laptops and SFF Desktops
NVIDIA is positioning RTX Spark as a premium, tightly controlled platform. Acer, ASUS, Dell, Gigabyte, HP, Lenovo, Microsoft, and MSI all plan RTX Spark systems, but the early laptops shown at Computex share similar aluminum chassis designs and tandem OLED displays, hinting that NVIDIA is dictating many design parameters. Microsoft’s Surface Laptop Ultra will serve as a flagship ARM Windows laptop based on RTX Spark, targeting 14‑ to 16‑inch form factors rather than entry-level devices. On the desktop side, vendors had small form factor RTX Spark PCs on stage, with designs that closely resemble existing GB10-based DGX Spark systems. This suggests highly integrated, low-volume desktops rather than large tower PCs with open customization. Together, these launches frame RTX Spark as an ARM Windows laptop and mini-desktop blueprint where NVIDIA controls thermals, performance targets, and the overall user experience.
Challenging x86: Software, Ecosystem, and What Comes Next
RTX Spark’s challenge to x86 is less about raw core counts and more about the ecosystem NVIDIA brings. CUDA, TensorRT, DLSS, Reflex, ray tracing, and the broader RTX software stack arrive intact on this ARM Windows laptop platform, sidestepping one of the biggest problems earlier Windows-on-Arm attempts faced: weak developer and application support. NVIDIA is light on benchmark data for now, and it has not compared RTX Spark directly against Intel, AMD, or Qualcomm chips, which keeps performance expectations uncertain. Still, with Blackwell GPU integration, Grace CPU cores, and unified memory already proven in GB10 systems, the consumer-focused N1X variant looks capable. A lower-end RTX Spark with about 400 TFLOPS of FP4 GPU performance is also on the roadmap, hinting at a family of Arm-based Windows PCs. The real test will be how well Windows scheduling, emulation, and native apps adapt to this new hybrid SoC baseline.
