What is NVIDIA RTX Spark?

At a glance: the current configurations

NVIDIA's current table labels both laptop tiers N1X. The flagship has a 20-core CPU, 6,144-core GPU and memory configurations reaching 128 GB; the second laptop tier has an 18-core CPU, 5,120-core GPU and a 64 GB ceiling. A product's actual memory, cooling, display and ports still come from its own SKU sheet, not the maximum platform configuration.

Vendor specifications, not measured wall power or a guarantee of sustained performance. Source: NVIDIA's current table ↗

Architecture and graphics: what the specifications mean

NVIDIA's CPU porting documentation identifies Arm Cortex-X925 and Cortex-A725 CPU cores. Arm describes the instruction-set environment; it does not, by itself, prove superior battery life against a particular x86 laptop. Battery capacity, screen settings, software execution path, drivers and the workload all affect an actual endurance test. NVIDIA CPU documentation ↗

The platform's Blackwell GPU provides RTX graphics capabilities. NVIDIA lists DLSS 5 and Reflex 2 in the current feature table. Availability still depends on application integration, drivers and settings. Matching the core count of a discrete GPU is not evidence of equal performance: memory behavior, clocks, power limits and cooling must also be considered.

For gaming, distinguish frames rendered by the game from generated frames. A demo described as smooth may be useful hands-on evidence without being a benchmark. Our benchmark evidence table separates leaked scores, historical engineering samples and vendor specifications; the reviews hub explains the limits attached to event and prototype coverage.

Unified memory and local AI

Unified memory gives the CPU and GPU a shared memory pool. A large installed capacity can be useful for creative assets and model weights, but an advertised capacity is not the amount available to a model after the operating system, application buffers and context cache are included. Lower-memory SKUs cannot be evaluated using the flagship's ceiling.

Peak AI arithmetic throughput is a manufacturer specification, not measured tokens/second or time-to-first-token. Native CUDA support also does not establish that every existing inference application's build and dependencies work unchanged on Windows Arm64. Start with the local AI testing guide: specify the model revision, quantization, context, backend, driver, power mode, quality and repeated results before comparing machines.

OEM partners and announced devices

ASUS, Dell, HP, Lenovo, Microsoft and MSI appear in our source-linked product catalog. ASUS and HP now have distinct laptop models rather than one combined rumor entry; compact desktops have different storage, networking and cooling priorities. The Lenovo convertible is tracked separately from Yoga Pro 9n. Manufacturer source conflicts and unverified fields stay visible on the individual pages.

• Microsoft Surface Laptop Ultra →Laptop · ANNOUNCED
• ASUS ProArt P14 (H7407) →Laptop · ANNOUNCED
• ASUS ProArt P16 (H7607) →Laptop · ANNOUNCED
• Lenovo Yoga Pro 9n (15-inch, Gen 11) →Laptop · ANNOUNCED
• Lenovo Yoga 9n 2-in-1 (16-inch, Gen 11) →Convertible · ANNOUNCED
• HP OmniBook Ultra 16 →Laptop · ANNOUNCED
• HP OmniBook X 14 →Laptop · ANNOUNCED
• HP OmniDesk with RTX Spark →Desktop · ANNOUNCED
• Dell XPS 16 Creator Edition →Laptop · ANNOUNCED
• ASUS ProArt GR1X →Desktop · ANNOUNCED
• MSI EdgeMesa N AI+ →Desktop · ANNOUNCED

Compare all device specifications and family guides →

Release timeline and regional availability

The Computex announcement established the platform. Subsequent manufacturer announcements and specification sheets make it possible to track actual product names and configurations. Neither stage establishes that a particular SKU has shipped, is available to preorder or is in stock at a local seller. We do not assign an unsupported country rollout wave or convert a global announcement into Canadian retail availability.

Check the country tracker and the product's linked regional manufacturer or retailer evidence before ordering. A launch window is not a promised delivery date. Prices, installed memory, storage options, warranty and return terms need to be verified against the exact local model identifier.

Compatibility and the buying decision

Windows on Arm supports native Arm64 applications and emulated x86/x64 applications, but those paths are not interchangeable with universal driver or plug-in support. Kernel components and anti-cheat systems need appropriate support. Check your actual games, peripherals and professional plug-ins, not just whether a demonstration launched successfully. Our compatibility checklist walks through that dependency chain.

If you need a machine for a deadline, compare devices that are demonstrably available and run your required software today. If you can wait, use independent retail tests to resolve the remaining performance and endurance questions. The Intel, AMD, Snapdragon and Apple comparison focuses on the evidence needed for a fair workload-specific decision, rather than declaring a platform-wide winner.