Nvidia Debuts RTX Spark for Windows PCs With 128GB Memory and 1 Petaflop AI
Updated
Updated · tech-insider.org · Jul 28
Nvidia Debuts RTX Spark for Windows PCs With 128GB Memory and 1 Petaflop AI
3 articles · Updated · tech-insider.org · Jul 28
Summary
RTX Spark marks Nvidia’s first direct push into Windows PC silicon, pairing a 20-core Grace CPU with a Blackwell GPU for laptops and compact desktops shipping this fall.
128GB of unified LPDDR5X memory is the headline feature, aimed at running large AI models locally by letting CPU and GPU share one memory pool instead of shuttling data between RAM and VRAM.
Nvidia claims 1 petaflop of FP4 AI performance and says the chip can handle models up to 120 billion parameters, though those figures have not yet been independently benchmarked.
Asus, Dell, HP, Lenovo, Microsoft Surface and MSI are launch partners, putting Nvidia into direct competition with Qualcomm, AMD, Intel and Apple in AI PCs; pricing was not disclosed.
The move extends Nvidia’s Grace-Blackwell strategy from data centers and the $3,999 DGX Spark desktop into mainstream PCs, betting buyers will pay for on-device AI despite rising memory costs.
Will TSMC's rising 3nm costs make Nvidia's new 128GB AI superchip too expensive for everyday Windows users?
Why is Nvidia repackaging data center tech into consumer laptops, and what does this mean for Intel's future?
Can Nvidia's unified memory truly run massive AI models locally, or will hidden bandwidth bottlenecks ruin the experience?
Nvidia RTX Spark Launch 2026: Arm Superchip Redefines Windows PCs with Local AI, Microsoft Partnership, and $2,000+ Pricing
Overview
Nvidia’s July 2026 launch of the Arm-based RTX Spark superchip marks a major shift in personal computing by moving AI processing from the cloud to local hardware, ensuring user privacy and eliminating ongoing cloud costs. Developed with MediaTek for top efficiency, the chip integrates CPU, GPU, and memory into a single SoC, but this design means users cannot upgrade components like RAM or graphics cards. To address security, Nvidia and Microsoft co-developed Windows security primitives and the OpenShell runtime, keeping sensitive data on-device and meeting strict regulations. However, legacy Windows software must run through Microsoft’s Prism emulation layer, which can reduce performance and battery life, making upgrades less practical for average users with standard workloads.