What Is the RTX Spark Grace CPU and Why It Matters
The RTX Spark Grace CPU is a 20‑core Arm processor that pairs high‑performance Cortex‑X925 cores with power‑efficient Cortex‑A725 cores to create a desktop AI platform using a smartphone‑style hybrid design. It sits at the heart of NVIDIA’s RTX Spark PC platform, alongside a Blackwell‑based RTX GPU and up to 128GB of unified LPDDR5X memory. Instead of separating system RAM and graphics VRAM, the entire platform shares one pool of memory, which helps AI models and traditional applications move data with less overhead. NVIDIA positions RTX Spark as the basis for thin laptops and compact desktops that still deliver serious graphics and AI performance. According to NVIDIA’s Computex presentation, RTX Spark can reach up to 1 PFLOP of AI performance in a design that device makers can fit into chassis as slim as 14mm.
Inside the Hybrid Core Architecture: Cortex-X925 and Cortex-A725
At the center of the RTX Spark Grace CPU is a hybrid core architecture built from 20 Arm CPU cores: ten Cortex‑X925 performance cores and ten Cortex‑A725 efficiency cores. The Cortex‑X925 cores handle demanding tasks such as compiling code, running complex AI models, or high‑frame‑rate gaming. The Cortex‑A725 cores focus on lighter or background workloads where low power draw matters more than peak speed. This arrangement mirrors modern smartphone chipsets, where big and small cores cooperate under a shared scheduler. In fact, the same Cortex‑X925 and Cortex‑A725 designs appear in recent phone platforms like MediaTek’s Dimensity 9400 and Dimensity 8500. By adopting this model, NVIDIA can push desktop‑class performance while improving energy efficiency compared with typical monolithic x86 CPUs that use identical cores for every type of workload.
Why Borrow Smartphone Chip Design for a Desktop AI Processor?
NVIDIA’s decision to base the RTX Spark Grace CPU on a smartphone‑style core mix comes down to efficiency and flexibility. Mobile chip designers have spent years refining ways to balance performance and battery life under tight thermal limits. By importing the big‑little concept into a desktop AI processor, NVIDIA gains fine‑grained control over power without sacrificing bursts of speed when needed. Light tasks can stay on Cortex‑A725 cores, keeping fans quiet and energy use low, while the Cortex‑X925 cores wake up for heavier work. This also helps thin laptops avoid throttling under sustained AI or gaming loads. Combined with unified LPDDR5X memory and the close CPU‑GPU link, the design tries to give PCs the responsiveness and all‑day efficiency of high‑end phones, while still supporting advanced RTX graphics features and Windows applications.
MediaTek, TSMC 3nm, and the Unified AI Platform
RTX Spark is a collaboration between NVIDIA and MediaTek, manufactured on TSMC’s advanced 3nm process. MediaTek contributed several key pieces: a custom memory controller for the unified LPDDR5X architecture, power management circuitry, and wireless connectivity blocks integrated directly into the chip. This tight integration supports low power draw even when the system is handling demanding workloads. The Grace CPU and Blackwell GPU connect over NVLink C2C with about 600GB/s bandwidth, allowing CPU and GPU to access the same memory pool without expensive copies. For AI, that means larger models and datasets can stay in place while different engines work on them. The result is a compact platform that blends a smartphone‑like SoC integration style with desktop‑class RTX graphics, aiming to power Windows PCs from thin laptops to small desktops later this year.
Flexible Workload Distribution for AI and Everyday Computing
The hybrid layout of the RTX Spark Grace CPU is designed for flexible workload distribution across AI and everyday desktop tasks. Heavy, latency‑sensitive jobs such as real‑time AI inference, complex simulations, or AAA gaming can span the ten Cortex‑X925 cores and the Blackwell GPU’s 6,144 CUDA cores and fifth‑generation Tensor Cores. Meanwhile, the ten Cortex‑A725 cores can keep background services, downloads, or office apps running with low impact on battery life and thermals. Unified memory ensures both CPU clusters and the GPU see the same data, which cuts down on duplication and delays when switching between tasks. This division of labor reflects how people use PCs today: long stretches of light work punctuated by intense sessions of AI, creation, or gaming that benefit from a smart, power‑aware core mix.
