What RTX Spark Is and Why NVIDIA Is Thinking in Generations
RTX Spark is an Arm-based superchip platform that combines a Grace CPU and Blackwell RTX graphics to power AI-heavy Windows PCs with unified memory and high bandwidth CPU–GPU links. Instead of treating RTX Spark as a one-off experiment, NVIDIA has positioned it as the first step in a multi-generation family of AI PC processors. Jensen Huang confirmed that the initial RTX Spark processor, internally called N1X, sits inside a roadmap that already includes N2X and N3X successors. He also revealed that “N1X is called N1X because it has a smaller version called N1,” hinting at future scaling options below the flagship part. That strategy signals an intent to build a lasting platform, where each new chip generation refines AI performance, power efficiency, and software support rather than resetting the design every few years.
Inside the First RTX Spark Processor: 3nm, Unified Memory and AI Throughput
At the heart of the first RTX Spark processor is a 3nm TSMC design that pairs a 20‑core Grace CPU with 6,144 CUDA cores and Blackwell RTX graphics. The CPU and GPU share up to 128GB of unified LPDDR5X memory across a high-speed NVLink C2C interconnect, which offers around 600GB/s of bandwidth between the two dies. This unified memory and fast link help RTX Spark reach an estimated 1 petaflop of AI compute performance for local workloads. For PC makers, the design targets thin laptops and compact desktops that must run AI agents, creator tools, and games without relying entirely on the cloud. According to NVIDIA’s Computex disclosures, these AI PC processors will ship in systems from Asus, Dell, HP, Lenovo, Microsoft Surface, MSI, and later Acer and Gigabyte, forming the first wave of RTX Spark-powered PCs.
Arm Cortex Architecture, MediaTek DNA and PC-Focused Core Tweaks
The Grace CPU inside RTX Spark uses an ARM Cortex architecture arranged in a mobile-style hybrid cluster. It combines ten high-performance Cortex-X925 cores with ten energy-efficient Cortex-A725 cores, similar to MediaTek’s Dimensity 9400 and Dimensity 8500 system-on-chips. MediaTek provided engineering and implementation expertise, helping NVIDIA adapt mobile core concepts to an AI PC processor with far higher sustained workloads. Die analysis shows that RTX Spark does not simply reuse phone silicon. The Cortex-X925 cores are smaller than in Dimensity 9400, but they adopt a power rail design akin to Dimensity 9500’s C1-Ultra, enabling higher sustained frequencies for demanding PC tasks. This focus on power delivery and scheduling helps RTX Spark stay within laptop thermal limits, even in designs like Microsoft’s Surface Laptop Ultra that run the chip at up to 110W TDP for extended periods.
From N1X to N2X and N3X: Building the AI PC Platform
With N2X and N3X already planned, NVIDIA is treating RTX Spark as the foundation of a long-term AI PC platform rather than a single generation of silicon. The current N1X design shows how mobile-derived ARM Cortex architecture can be tuned for PC workloads; future parts are likely to refine that approach with improved AI engines, more memory bandwidth, and potentially broader form factors, from compact desktops to gaming handhelds. Huang has compared RTX Spark-powered systems to devices people keep for five to ten years, suggesting a roadmap that supports long software lifecycles and continuous AI model evolution. As anti-cheat support and Windows ecosystem compatibility mature, the N2X and N3X chips are expected to push RTX Spark deeper into mainstream consumer PCs, turning local AI agents and always-on assistants into standard features rather than niche extras.
