What RTX Spark and Its Hybrid Grace CPU Actually Are
NVIDIA RTX Spark is an Arm-based PC platform that combines a custom 20-core Grace CPU, a Blackwell RTX GPU, and unified memory into a single chip designed to run AI and everyday PC workloads efficiently. At the heart of RTX Spark’s CPU is a mobile-inspired hybrid processor design: ten Cortex-X925 performance cores paired with ten Cortex-A725 efficiency cores. This layout mirrors modern smartphone chips, but it is tuned for laptops and compact desktops instead of phones. The Cortex-X925 cores provide the heavy lifting for AI inference, gaming, and compilation, while the Cortex-A725 cores handle lighter background tasks and idle work. By mixing these core types and scheduling tasks between them, RTX Spark aims to stretch battery life without giving up the high frame rates and AI responsiveness users expect from an RTX-powered PC.
Inside the RTX Spark CPU Architecture: Mobile DNA at 3nm
The RTX Spark CPU architecture is built on TSMC’s 3nm process and uses Arm’s latest Cortex designs as its foundation. NVIDIA and MediaTek worked together to adapt these mobile-class cores for an ARM architecture desktop platform, pairing the 20-core Grace processor with 6,144 CUDA cores and Blackwell graphics. According to NVIDIA, this unified design can access up to 128GB of LPDDR5X memory over NVLink C2C at about 600GB/s, helping the system reach an estimated 1 PFLOP of AI performance. MediaTek’s role goes beyond the CPU cores: it designed the custom memory controller, power management circuitry, and wireless components, all drawn from its experience with Dimensity smartphone chipsets. The result is a PC SoC that borrows the tight integration and power efficiency of phones while scaling up to workstation-like AI performance in thin laptops and small desktops.
How Cortex-X925 and A725 Cores Split Work for AI and Everyday Tasks
RTX Spark’s Grace CPU relies on a heterogeneous layout: ten Cortex-X925 performance cores and ten Cortex-A725 efficiency cores share the workload. The Cortex-X925 cores, familiar from MediaTek’s Dimensity 9400 family, are built for demanding tasks such as real‑time AI processing, 3D rendering, and heavy multitasking. In contrast, Cortex-A725 efficiency cores, similar to those in the Dimensity 8500, take over background services, web browsing, or office work where peak speed matters less than battery life. This dynamic switching is central to the hybrid processor design. The scheduler can wake powerful cores only when an application needs them and otherwise prefer the efficiency cores, reducing heat and fan noise. For AI workloads, this means bursts of high performance when models are running, followed by cooler, quieter operation when users move back to lightweight tasks.
Tweaked Cortex-X925 Cores: From Phone Silicon to PC Workhorse
Although RTX Spark uses Cortex-X925 cores similar to those in MediaTek’s Dimensity 9400, they are not simple copies. Die analysis shows the Grace CPU’s Cortex-X925 blocks are smaller and integrate a power rail design borrowed from the Dimensity 9500’s C1-Ultra, supporting higher sustained frequencies for PC workloads. This matters for AI and content creation, where long multi-core sessions are common. The efficient power distribution and scheduling logic, inherited from Dimensity 9500 designs, help RTX Spark maintain clocks without quickly hitting thermal limits. Early demonstrations include laptops such as Microsoft’s Surface Laptop Ultra running the SoC at up to 110W TDP while keeping performance stable. These alterations highlight how a single Arm core design can be tuned differently for mobile SoCs and for an ARM architecture desktop platform focused on continuous, heavy workloads.
Why Mobile-Style Cores Improve PC AI Performance
For AI-focused PCs, the benefit of RTX Spark’s mobile-inspired layout is not only raw speed but how efficiently the entire system moves data and power. Unified LPDDR5X memory means large AI models do not need to bounce between separate CPU and GPU pools, cutting latency and boosting throughput. At the same time, efficiency cores keep background operations from wasting power on the big cores, leaving more thermal and power headroom for AI bursts on the Cortex-X925 and the Blackwell GPU. NVIDIA’s RTX software stack—CUDA, TensorRT, DLSS, Reflex, and ray tracing—sits on top, turning this CPU–GPU–memory foundation into a cohesive AI PC platform. Hybrid cores from the smartphone world, tuned power rails, and unified memory together make RTX Spark’s Grace CPU architecture a bridge between mobile efficiency and desktop-grade AI performance.
