What Is RTX Spark and Its 20-Core Grace CPU?
The NVIDIA RTX Spark processor is an Arm-based system-on-chip for Windows PCs that combines a 20-core Grace CPU, a Blackwell GPU, and up to 128GB of unified LPDDR5X memory into a single platform designed for high-end laptops and compact desktops. At the heart of RTX Spark sits a custom Grace CPU architecture developed with MediaTek, using ten Cortex-X925 cores for peak performance and ten Cortex-A725 cores for energy-efficient tasks. This hybrid core layout mirrors modern smartphone chipsets, but scaled up for PC-class workloads and cooling envelopes. RTX Spark is NVIDIA’s second attempt at Windows-on-Arm, and it stands out because it brings the full RTX software stack—CUDA, TensorRT, DLSS, Reflex, and ray tracing—to an Arm SoC Windows platform. For users, that means familiar GeForce-style features on a new CPU architecture that aims to balance speed, battery life, and AI throughput.
Inside the Grace CPU Architecture: Cortex-X925 and Cortex-A725
Grace CPU architecture in RTX Spark follows a big-and-small core design similar to premium phone processors, but with desktop ambitions. NVIDIA and MediaTek’s 20-core layout pairs ten high-performance Cortex-X925 cores with ten power-saving Cortex-A725 cores. The X925 cores are tuned for heavy workloads such as gaming, code compilation, and complex simulations, while the A725 cluster takes over background tasks, light browsing, and idle states. According to NVIDIA’s published specifications, this configuration “should look familiar to anyone following the smartphone chipset market,” underlining how closely it tracks mobile design principles. What changes in the PC context is thermal headroom and sustained performance: more space and better cooling allow the Cortex-X925 cores to run at higher sustained clocks, helping RTX Spark bridge the gap between traditional x86 CPUs and mobile-first Arm SoC Windows designs in demanding applications.
Hybrid Design: Borrowed from Smartphones, Tuned for PCs
The hybrid Grace CPU configuration in RTX Spark borrows heavily from smartphone design philosophy, but the goals stretch beyond saving battery on a handheld device. In phones, Arm big-and-small layouts keep heat and power in check under tight constraints; NVIDIA is now applying the same logic to Windows laptops and compact desktops. The operating system can schedule bursty, latency-sensitive tasks to Cortex-X925 cores while offloading low-priority work to Cortex-A725 cores, lowering average power draw and fan noise. This can help thin-and-light RTX Spark laptops deliver all-day productivity without sacrificing short, high-performance bursts. At the same time, the shared LPDDR5X memory pool and close coupling between CPU and GPU reduce data movement overhead. For typical Windows users, that should translate into faster wake times, smoother multitasking, and more consistent performance under mixed workloads like streaming, browsing, and office apps running alongside AI tools.
Blackwell GPU and Unified Memory: AI and Graphics on One Chip
Beyond the Grace CPU, RTX Spark integrates a Blackwell GPU with 48 streaming multiprocessors and 6,144 CUDA cores, bringing desktop-class graphics into an SoC form factor. NVIDIA claims up to 1 PFLOP of AI performance using its fifth-generation Tensor Cores, a figure comparable to mid-range desktop GPUs. The entire chip is fed by up to 128GB of unified LPDDR5X memory, meaning CPU and GPU share the same pool instead of separate system and graphics RAM. According to NVIDIA’s own platform description, this unified memory design can be as important as raw compute for AI workloads, because large models no longer need to fit within a discrete VRAM limit. For Windows users running generative AI, video editing, or 3D rendering, the combination of Grace CPU architecture and a Blackwell GPU inside a single RTX Spark processor promises shorter load times, fewer out-of-memory errors, and smoother real-time previews.
What RTX Spark Means for Windows-on-Arm Users
RTX Spark signals a serious push to make Arm SoC Windows machines feel like full GeForce PCs rather than experimental devices. Systems based on RTX Spark are planned for premium 14- to 16-inch laptops with aluminum chassis and OLED displays, as well as small form factor desktops that resemble NVIDIA’s existing GB10 designs. Because RTX Spark mirrors the GB10 SoC in core counts, Blackwell GPU configuration, and memory capacity, its performance profile is not entirely unknown to data center customers already using DGX Spark machines. The difference is that RTX Spark is built to run Windows, complete with CUDA, DLSS, and ray tracing support. For users choosing between x86 and Arm, this means a new class of Windows-on-Arm devices that can run demanding games and AI-enhanced applications, while still benefiting from the efficiency and instant-on behavior associated with mobile-derived architectures.
