Nvidia is targeting a repositioning of the personal computer market with the introduction of RTX Spark, a new Arm-based system-on-chip (SoC) platform that integrates the company’s Blackwell architecture into thin and light Windows laptops and mini desktops. The chip aims to deliver high-performance processing for personal AI agents, creative applications, and gaming, while minimizing the space, power consumption, and cooling demands typically associated with discrete graphics solutions.
RTX Spark enters the Windows on Arm ecosystem alongside Qualcomm’s Snapdragon X processors, which similarly promise all-day battery life. However, Nvidia positions the Spark for substantially heavier computational workloads than what Snapdragon processors are designed to handle, including rendering ultralarge 90GB-plus 3D scenes, editing 12K 4:2:2 video, generating 4K AI videos, running 120B-parameter large language models with up to 1 million tokens context using local agents, and playing AAA games at 1440p and over 100 frames per second.
The initial RTX Spark models are scheduled to launch this fall across multiple manufacturers and form factors, including the Microsoft Surface Laptop Ultra, Dell XPS 16, Asus ProArt P14 and P16, HP Omnibook X 14 and Omnibook Ultra 16, Lenovo Yoga Pro 9i, and MSI Prestige N16 Flip AI.
The 15-inch Surface Laptop Ultra stands out particularly due to Microsoft’s prolonged neglect of touchscreen upgrades in the Surface lineup, as well as the absence of discrete GPU integration in previous models despite premium pricing. The Surface Laptop Ultra introduces a higher-resolution (262ppi) 15-inch mini LED touchscreen supporting HDR with peak brightness of 2,000 nits—marking a significant improvement over older generations.
Nvidia’s strategy includes a complete product line spanning desktop, laptop, and workstation segments, with mini desktops expected from manufacturers such as Acer, Asus, Dell, HP, and Lenovo. These Spark-based systems will compete directly with AMD’s Ryzen AI Halo-based offerings and Apple’s M5 Pro and M5 Max MacBook Pros.
The RTX Spark is derived from Nvidia’s DGX Spark (GB10) platform, originally designed for developer-focused Linux desktops and Windows-based DGX Stations. Developed in collaboration with MediaTek, the chip features 6,144 CUDA cores, a 20-core Grace CPU, and supports up to 128GB unified memory. Nvidia states it can handle 120B parameter AI models with a 1 million token context window.
The RTX Spark under the hood of the Surface Laptop Ultra.
While GPU performance is comparable to an RTX 5070, the unified memory architecture provides significantly more RAM access than traditional 12GB VRAM configurations. System memory configurations may start as low as 16GB, potentially creating bottlenecks under certain workloads compared to discrete GPU implementations. Nvidia demonstrated 100fps at 1440p resolution, though it was unclear whether this included DLSS 4.5 acceleration.
The chip achieves approximately 1 petaflop of AI performance using FP4 calculations—a format currently favored for its speed advantages over other floating-point representations, despite some accuracy trade-offs. This marks the first consumer SoC natively supporting FP4 in hardware.
This illustration of the RTX Spark in situ has the fuzzy, glowy look of a generated image.
Power consumption varies significantly, ranging from single-digit watts to 80W, meaning buyers must carefully evaluate whether specific laptop models operate at peak performance or are subject to manufacturer throttling. The adaptable power profile introduces variability in real-world performance, particularly on battery power.
To qualify as Copilot Plus-certified systems, RTX Spark devices must include an NPU capable of delivering at least 40 TOPS (trillion operations per second). Nvidia has collaborated extensively with Microsoft to optimize Windows for the new architecture, including updates to the Prism emulation layer that previously supported only Qualcomm’s SoCs.
Windows modifications focus on efficiently distributing workloads across CPU cores, balancing thermal management with performance, and intelligently handling unified memory allocation for AI processing through TensorRT. Nvidia has also worked with Microsoft to improve compatibility with gaming-related anti-cheat software and the Xbox application, supporting Microsoft’s cross-platform gaming strategy.
Software partners like Adobe are adapting their imaging engines to leverage RTX Spark directly, with new acceleration pipelines for GPU- and AI-intensive features in applications such as Premiere Pro and Photoshop. Additionally, Nvidia is porting its OpenShell security protocols—which enable secure AI agent deployment through configurable guardrails, local model routing, and privacy preservation for cloud-based queries—to Windows via controls to be revealed at Microsoft’s Build conference in June.
Nvidia’s vision extends beyond developer markets, aiming to expand AI agent adoption among general users by addressing security and privacy concerns around running agents on primary PCs. OpenShell will integrate with existing agent frameworks including OpenClaw and Hermes.
Supply chain pressures stemming from AI’s voracious demand for components—including memory, processors, and SSD storage—have created significant shortages, driving up computer and smartphone prices while limiting configuration availability. Final pricing for RTX Spark systems will be announced closer to their fall availability.
Current market dynamics suggest RTX Spark systems may carry premium pricing due to component scarcity, though specific configurations and cost structures remain undetermined pending manufacturer announcements and shipping preparations.