What RTX Spark Is: A Hybrid ARM Superchip for PCs
NVIDIA RTX Spark is a 3nm ARM-based superchip platform for Windows PCs that combines a hybrid Grace CPU complex with Blackwell RTX graphics and unified LPDDR5X memory to deliver desktop-class performance with mobile-grade power efficiency. At the heart of the RTX Spark processor is a 20-core Grace CPU built on ARM Cortex architecture, paired with 6,144 CUDA cores and next‑generation Blackwell graphics. The design follows a mobile-style big.LITTLE layout, with ten high‑performance Cortex‑X925 cores and ten efficient Cortex‑A725 cores, similar to those used in MediaTek’s Dimensity 9400 and 8500. NVIDIA connects CPU and GPU through NVLink C2C, offering around 600 GB/s bandwidth and enabling up to an estimated 1 petaflop of AI compute in a single package. This unified approach aims to rethink what an NVIDIA desktop CPU can look like.
Altered Cortex‑X925 Cores: From Phone SoCs to PC Workhorses
Although RTX Spark borrows its Cortex‑X925 DNA from MediaTek’s Dimensity 9400, the cores are not drop‑in copies. Die analysis shared by Geekerwan shows that RTX Spark’s X925 cores are smaller than those in the earlier mobile SoC and adopt the power rail design from the Dimensity 9500’s C1‑Ultra cluster. According to Wccftech’s reporting on Geekerwan’s findings, this altered layout is meant to sustain higher clock speeds for demanding PC workloads without crossing thermal limits. That matters in designs such as the Surface Laptop Ultra, where RTX Spark configurations with 110 W TDP need long‑running performance rather than short mobile‑style bursts. By combining attributes from both Dimensity 9400 and 9500, NVIDIA and MediaTek turn a smartphone‑class core into a desktop‑ready building block for the RTX Spark processor.
Hybrid Core Layout Meets Unified 3nm Chip Design
The RTX Spark processor extends mobile CPU ideas into a full desktop platform by pairing its hybrid core layout with a unified 3nm chip design. The ten Cortex‑X925 performance cores handle high‑thread, latency‑sensitive jobs such as compiling, complex simulations, and game logic, while the ten Cortex‑A725 efficiency cores take on background tasks and light parallel work. This keeps overall power draw closer to a mobile SoC profile, even as the GPU and AI blocks ramp up. Manufactured on TSMC’s 3nm node, the chip connects Grace CPU and Blackwell GPU using NVLink C2C, with roughly 600 GB/s bandwidth and shared access to up to 128 GB of LPDDR5X. The result is a desktop ARM Cortex architecture that behaves more like a single super‑accelerator than a loose CPU‑GPU pairing on a motherboard.
MediaTek’s Role: Memory, Power, and Radio for a PC‑Class SoC
MediaTek’s contribution goes far beyond licensing cores. For RTX Spark, MediaTek engineered the custom memory controller driving the unified LPDDR5X pool, designed the full power‑management circuitry, and integrated the wireless transmission blocks directly into the SoC. That combination aims to keep idle and light‑load draw low while allowing aggressive power delivery when the 20 cores and 6,144 CUDA cores are active. In effect, MediaTek brings its experience building tightly integrated mobile SoCs into the NVIDIA desktop CPU space. NVIDIA, in turn, aligns its Grace and Blackwell IP around this mobile‑inspired foundation, producing a superchip that can be dropped into thin‑and‑light laptops as well as RTX Spark‑based desktops. The partnership signals a long‑term plan, with Wccftech noting expectations for more advanced RTX Spark versions arriving in 2027.
Why RTX Spark Matters for ARM‑Based Desktop Computing
RTX Spark is more than another NVIDIA processor; it is a template for ARM‑based desktop computing that blurs the traditional lines between phone SoCs and PC platforms. By proving that altered Cortex‑X925 cores, derived from Dimensity 9400 and 9500 designs, can sustain high PC‑class frequencies, NVIDIA shows that mobile‑inspired ARM cores can step into the role of a primary desktop CPU. At the same time, the 3nm chip design, unified memory, and NVLink C2C link turn the superchip into a tightly coupled CPU‑GPU‑AI engine that resembles NVIDIA’s DGX Spark layout at a smaller scale. For developers and system designers, this means treating GPU, CPU, and memory as parts of one coherent RTX Spark processor rather than separate components, with potential gains in AI throughput, power efficiency, and form‑factor flexibility.
