Table of Contents
NVIDIA announced the RTX SparkA new superchip for Windows notebooks and compact desktops promises to bring AI agents, advanced content creation, and gaming to local processing, without always relying on the cloud. The platform combines a Blackwell RTX GPU, Grace CPU, and up to 128 GB of unified memory, with its arrival expected in the Northern Hemisphere autumn in models from manufacturers such as ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI.
What is NVIDIA RTX Spark?
The RTX Spark is NVIDIA's new offering for a PC category aimed at... IA localInstead of being just a graphics card or a conventional processor, the proposal is to bring together in a single package the AI, graphics, and accelerated computing features that the company already uses in RTX GPUs and in its software stack for developers.
According to NVIDIA, the chip features a Blackwell RTX GPU with 6.144 CUDA cores.Fifth-generation Tensor Cores with FP4 precision and a NVIDIA Grace 20-core CPU, connected by NVLink-C2C. The CPU was developed in collaboration with MediaTek, a move that reinforces NVIDIA's attempt to gain more space in the heart of PCs, where Intel, AMD, Qualcomm, and Apple already compete with their own solutions.
RTX Spark is a superchip, not just a CPU.
On a conventional computer, the CPU (Central Processing Unit), or simply processorThe chipset is the main system chip, responsible for executing program instructions, managing operating system tasks, and coordinating other components.
Already GPU (Graphics Processing Unit) is a separate chip specialized in massive parallel computations, making it much more efficient at rendering images, processing visual effects, and running artificial intelligence models.
Already RTX Spark is an superchip: a single-chip design that combines CPU and GPU in a single package. On one side, the Blackwell RTX GPU with 6.144 CUDA cores and fifth-generation Tensor Cores with FP4 precision. On the other hand, the 20-core Grace CPU, developed in collaboration with MediaTek.
One of the most practical consequences of this integrated architecture is... unified memoryInstead of the CPU and GPU having separate memory pools, they share the same memory pool. 128GB LPDDR5XThis allows large AI models and heavy 3D scenes to be fully loaded into memory without transfers between chips, reducing bottlenecks and increasing overall system efficiency.
Local AI: Why does it matter?
The main promise is to allow AI agents to work directly on the user's computer. In the vision advocated by Jensen Huang, CEO of NVIDIA, the PC ceases to be merely a tool operated by clicks, menus, and isolated applications, and begins to work as an "assistant" capable of performing tasks between programs, searching local files, generating media, and automating workflows.
To that end, the company claims that the RTX Spark can offer Up to 1 petaflop of AI performance. e Up to 128 GB of unified memory.This combination would allow running language models locally with up to 120 billion parameters and context of up to 1 million tokens, as well as multimodal flows for creating images, video, and code.
Running models on the device itself can bring three practical advantages: lower latency, greater privacy, and functionality even when the internet connection is limited. However, this approach doesn't eliminate the cloud: NVIDIA mentions intelligent routing between local and online models, depending on the task, privacy policies, and device capabilities.
Partnership with Microsoft and security for agents.
The new feature was announced in conjunction with Microsoft, which is working with NVIDIA on Windows features for personal agents. The foundation includes new system security primitives, with identity, containment, policies, and permission control, in addition to... NVIDIA OpenShell, a runtime designed to define what agents can and cannot do on the PC.
In practice, the idea is to prevent an agent from having unrestricted access to sensitive files, applications, or data. OpenShell also promises to direct queries to local models when privacy is a priority and mask personal information when it is necessary to resort to cloud models. This point is essential because autonomous agents will only be useful on the user's main computer if there is clear control over permissions and actions performed.
Creation, games and compatible software
Despite its focus on AI agents, the platform also targets creators, developers, and gamers. NVIDIA cites rendering 3D scenes above 90 GB with OptiX and DLSS, 12K 4:2:2 video editing using the Blackwell decoder, AI-powered 4K video generation, and AAA gaming at 1440p above 100 frames per second with ray tracing, DLSS, and Reflex.
Regarding performance in games, the Videocardz According to information obtained from NVIDIA, the RTX Spark can achieve graphics performance close to that of a... GeForce RTX 5070 Laptop GPU, depending on the application.
The company cited Indiana Jones and the Great Circle It was listed as one of the compatible titles, but did not disclose a complete list of games, graphics settings used in the tests, or power consumption limits to support the promise of 1440p above 100 FPS.
Windows limitations on ARM
It is worth noting that, being an architecture-based platform ARMHowever, RTX Spark has one limitation to consider: traditional Windows applications, developed for processors. x86They may need to be run through an emulator. Prism From Microsoft if native ARM versions are not available.
This compatibility is not yet universal, but it has been improving consistently as the ARM ecosystem grows on Windows. This includes the arrival of a platform the size of... RTX Sparkwith names such Adobe, Blender, DaVinci Resolve e CapCut Already developing native versions, it can further accelerate this transition in the coming years.
With that in mind, the company stated that Over 100 software vendors and studios support RTX Spark., including Adobe, Blackmagic Design, Blender, CapCut, ComfyUI, OTOY, KRAFTON, NetEase, Remedy, Riot Games and Xbox. Adobe Premiere e Photoshop They will also be restructured to take advantage of the platform, promising up to 2x greater performance in AI, editing, color grading, and effects.
When will the first PCs with RTX Spark arrive?
The first thin notebooks and compact desktops with RTX Spark are expected to be released soon. autumn in the northern hemisphere, a period between September and November. The new Surface Ultra It has already been announced as the first to feature the new superchip and ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI are in the first wave. while Acer and Gigabyte are expected to release models later.
NVIDIA has not yet announced pricing, final configurations by manufacturer, or availability in Brazil. Therefore, for the Brazilian consumer, RTX Spark should initially be viewed as a next-generation technology for high-performance PCs, closer to portable workstations than conventional notebooks.
What changes for the user?
If the proposal materializes in real products, RTX Spark could accelerate a significant shift in the computer market: the transition from "AI-enabled PCs" to machines capable of running complex assistants on the device itself. This is of interest to those who work with automation, development, video editing, 3D modeling, image generation, gaming, and local search across large volumes of files.
The challenge will be to transform the promise into a simple and secure experience. Raw performance, unified memory, and software support are important parts of the equation, but adoption will depend on price, battery life, availability, agent quality, and user confidence in allowing AI to operate within Windows.
Watch the videos below to learn more:
And you? What did you think of the announcement? RTX SparkAre you interested in trying out an ARM-based switch optimized for local AI agents? Leave your feedback in the comments. Comment and on social networks showmetech!
See also:
Text revised by Diego Amorim in 01 / 06 / 2026.
Sources: Videocardz, NVIDIA Newsroom, NVIDIA DGX Spark e Android Authority.
Discover more about Showmetech
Sign up to receive our latest news via email.