NVIDIA RTX Spark launches on new Windows PCs with 128 GB of unified memory

A new category of portable PC

NVIDIA has launched the RTX Spark platform, which is designed to bring high-end AI computing to portable Windows devices. The project is a collaboration between NVIDIA, Microsoft, and several major computer manufacturers. These new machines aim to balance the power of a workstation with the portability of a laptop.

The hardware is built around a specialized superchip that integrates an NVIDIA Grace CPU and an NVIDIA Blackwell RTX GPU. This combination allows the device to handle demanding tasks like 3D rendering and AI model management. The goal is to provide a versatile platform for creators, developers, and gamers who need to work away from a power outlet.

Several major brands are offering these new devices, including the Dell XPS 16, HP OmniBook Ultra, and Microsoft Surface Laptop Ultra. The first models are available worldwide starting at $2,599. While prices vary by configuration, the platform is intended to compete in the high-end portable market.

How unified memory works

One of the defining technical features of RTX Spark is its unified memory architecture. In this setup, the CPU and GPU share the same pool of resources, which provides greater flexibility for demanding workloads. This allows the system to manage advanced AI models and complex 3D scenes that exceed 90 GB in size.

By sharing memory, the system can quickly switch between different creative tasks and software development environments. This architecture is specifically designed to handle heavy assets without the bottleneck of moving data between separate memory pools. It enables the device to run large-scale AI models directly on the local hardware.

The platform also includes hardware acceleration for video and various technologies from the NVIDIA ecosystem. These include CUDA for compute power, DLSS for image quality, and TensorRT for AI performance. These tools work together to ensure the device remains performant even when running high-resolution content.

Local AI agents and tools

NVIDIA RTX Spark is designed to move beyond simple chatbots by supporting local AI agents capable of completing multi-step tasks. These agents can generate content, write code, and automate workflows without needing an internet connection. To make this accessible, NVIDIA is providing guided procedures that do not require using a command line.

The platform supports a variety of existing tools, including GitHub Copilot, Perplexity, and WorkBuddy. For those running local models, it supports the Nemotron 3.5 Bolt 74B-A9B model, which is optimized for tool use. This allows users to keep their code and context on their own devices, which can improve privacy and reduce latency.

Security is also a focus, with NVIDIA OpenShell and Microsoft Execution Containers managing permissions for these agents. Furthermore, NVIDIA PAIR allows a single RTX Spark device to share its AI resources with other computers on a home network. This means one machine can handle the heavy AI workload while other PCs remain free for gaming or creative work.

Impact on creative workflows

Creative professionals can use the new platform to edit 8K video in 4:2:2 format, even when the laptop is not plugged into a power source. Adobe is integrating NVIDIA TensorRT acceleration into several of its flagship applications, including Photoshop and Premiere. These updates improve various stages of production, such as rendering, color grading, and video playback.

For 3D artists, the platform supports heavy scenes and provides tools for converting AI-generated videos into smoother content. For example, users can convert 15 FPS AI videos into 60 FPS content without relying on paid cloud processing. This is particularly useful for creators who want to maintain high-quality output while keeping costs down.

Video creators can also use NVIDIA Broadcast 2.3 to remove background noise from recorded footage. Additionally, the RTX Video beta features can improve footage through super-resolution up to 4K. These tools are designed to maintain high performance in thin and portable devices without sacrificing autonomy.

Gaming and developer support

For gamers, the Blackwell RTX GPU provides technologies like ray tracing and G-SYNC to enhance visual quality. NVIDIA has showcased the platform with titles like Cyberpunk 2077 and Fortnite. The hardware supports DLSS 5 and 3D-Guided Neural Rendering to improve lighting and material details in real-time.

Developers also receive significant updates, particularly for those working on Windows with Arm architecture. NVIDIA has released a first version of PyTorch with CUDA for Windows on Arm and a native Python distribution. The platform claims compatibility with 97% of the 15,000 most commonly used Python packages.

The ultimate goal is to help developers create NVIDIA-accelerated applications that can be distributed across a massive network of GPUs. By combining code writing, testing, and prototyping on a single machine, the RTX Spark platform aims to streamline the entire development lifecycle. This creates a bridge between local development and large-scale cloud deployment.