Nvidia unveils RTX Spark superchip for AI-focused Windows PCs: Details

The Nvidia RTX Spark superchip combines its Grace CPU, Blackwell RTX GPU, and up to 128GB unified memory, targeting developers, creators and gamers

Nvidia has unveiled RTX Spark, a superchip that it says is designed to power a new generation of Windows laptops and compact desktop PCs built for AI, gaming, and content creation. Announced at Computex 2026 trade show, RTX Spark is being positioned by the company as hardware for what it calls the era of personal AI agents, enabling advanced AI workloads to run directly on local devices.

 

According to Nvidia, the first RTX Spark-powered Windows PCs will be available this fall from manufacturers including Microsoft, ASUS, Dell, HP, Lenovo and MSI, with Acer and GIGABYTE expected to introduce devices later.

What is Nvidia RTX Spark

According to Nvidia, RTX Spark superchip combines a 20-core Nvidia Grace CPU with a Blackwell RTX GPU featuring 6,144 CUDA cores and fifth-generation Tensor Cores with FP4 precision. The CPU and GPU are connected using Nvidia’s NVLink-C2C chip-to-chip interconnect.

 

The company said the superchip can deliver up to 1 petaflop of AI performance and support up to 128GB of unified memory. Unlike conventional PCs where the processor and graphics chip typically access separate memory pools, unified memory allows both components to share the same memory resources, which Nvidia says can benefit AI, graphics and creator workloads.

 

Nvidia also revealed that MediaTek collaborated on the custom CPU design used in RTX Spark, contributing to its power efficiency, performance and connectivity capabilities.

What can RTX Spark do

Nvidia is positioning RTX Spark as a single system for AI development, content creation and gaming. According to the company, the superchip brings together technologies including CUDA, RTX, DLSS, TensorRT, OptiX, Reflex and G-SYNC in thin-and-light laptops and compact desktop PCs.

 

The company claims RTX Spark can run large language models with up to 120 billion parameters locally while supporting context windows of up to one million tokens. Nvidia said the hardware is designed to provide enough computing power and memory for increasingly capable AI workloads without relying entirely on cloud infrastructure.

 

For creators, Nvidia said RTX Spark can render 90GB-plus 3D scenes, edit 12K 4:2:2 video using Blackwell media engines, and generate 4K AI videos locally. The company also highlighted support for professional content creation applications and AI-assisted workflows.

 

On the gaming side, Nvidia claims RTX Spark systems will be capable of running AAA games at 1440p resolution and more than 100 frames per second using technologies such as ray tracing, DLSS and Reflex. Nvidia also announced DLSS 4.5 Ray Reconstruction, which it says will improve image quality in ray-traced and path-traced games. The technology is expected to arrive in supported games and Blender 5.3 later this year.

Focus on AI agents

A major focus of Nvidia’s announcement is support for AI agents that can perform tasks directly on users’ devices. According to the company, RTX Spark has been designed to run AI agents locally while maintaining privacy and user control. To support this, Nvidia introduced OpenShell, a new runtime that helps developers build and deploy AI agents on Windows.

 

Nvidia said OpenShell allows users to define what agents can access, apply privacy policies, route requests between local and cloud-based AI models, and help protect personal information before data is sent to external services. The company added that developers behind projects such as Hermes Agent and OpenClaw are already integrating OpenShell into their Windows applications.

 

According to Nvidia, these agents could eventually help users perform tasks across Windows applications, automate workflows, search files, generate content and assist with software development while running directly on local hardware.

Software and app support

Nvidia said more than 1,000 RTX-enhanced and accelerated games and applications are already available across its ecosystem. The company also highlighted support from a wide range of software developers for RTX Spark-powered systems.

 

According to Nvidia, more than 100 software companies, including Adobe, Blackmagic Design, Blender, CapCut, ComfyUI and OTOY, are supporting the platform. The company said RTX Spark’s combination of AI performance and large unified memory is intended to benefit workloads such as video production, 3D rendering, AI image generation and local model inference.

 

Nvidia said Adobe is rearchitecting Photoshop and Premiere for RTX Spark, enabling the applications to make deeper use of the superchip’s Blackwell GPU, unified memory architecture and TensorRT software stack.

 

As per the chipmaker, the changes are expected to improve AI-powered editing, rendering, colour grading and graphics performance. Adobe also plans to extend Photoshop and Premiere with support for Windows-based AI agents that can assist users during creative workflows.

What Microsoft is adding

Alongside Nvidia’s announcement, Microsoft said it has worked closely with Nvidia to optimise Windows for RTX Spark-powered systems. According to Microsoft, RTX Spark devices will run Windows on Arm and include several platform-level optimisations for the new hardware. The company said it has tuned Windows scheduling to better distribute workloads across RTX Spark’s CPU architecture and implemented its Microsoft Power and Thermal Framework to improve performance and efficiency.

 

Microsoft also noted improvements for unified memory systems, including increasing the amount of system memory accessible to the GPU and enhancing memory management for demanding AI, creator and gaming workloads.

 

For compatibility, Microsoft confirmed that Prism, its emulator for running 32-bit and 64-bit x86 applications on Windows on Arm, will be available on RTX Spark systems. The company said Prism has been specifically tuned for RTX Spark and includes compatibility improvements introduced over the past year.

 

Microsoft further said RTX Spark systems will support Windows ML and native TensorRT integration, allowing developers to access Nvidia AI technologies directly through Windows.

 

The company also highlighted broad software support for Windows on Arm, including applications such as Blender, DaVinci Resolve, Maxon Cinema 4D, Maxon Redshift, Topaz Photo, CapCut, Cubase, Bitwig Studio, Affinity by Canva and MATLAB. Microsoft added that Photoshop and Premiere already run natively on Arm and are receiving additional optimisations through collaboration with Adobe and Nvidia.

Which devices will use RTX Spark

The first RTX Spark-powered devices are expected to launch this fall. Microsoft has already announced the Surface Laptop Ultra, a new Surface device aimed at creators, developers and professionals working with demanding AI and content creation workloads.

 

Other manufacturers including ASUS, Dell, HP, Lenovo and MSI are also in pipeline to introduce RTX Spark-powered systems. According to Nvidia, these will include thin-and-light laptops as well as compact desktop PCs.

 

Nvidia said RTX Spark-powered laptops will be available in 14-inch to 16-inch sizes, with some designs being 14mm thick and weighing around three pounds. The company also said the devices will feature premium aluminium chassis designs and tandem OLED displays with G-SYNC support.