NVIDIA Launches RTX Spark Windows PCs for Local Frontier AI Computing
Nvidia and PC makers roll out Grace Blackwell laptops with up to 128GB unified memory, targeting local AI agents, content creation, and gaming.
In short
- RTX Spark laptops integrate an Nvidia Grace CPU and Blackwell RTX GPU on a single SoC with up to 128GB of unified memory and 1 petaflop of FP4 compute.
- Launch partners including ASUS, Dell, HP, Lenovo, Microsoft, and MSI released systems on October 16, 2026, starting at $2,599, with compact desktops arriving in November.
- The platform targets local AI agents and developers with native CUDA support, a dedicated 40 TOPS NPU for Copilot+, and tools like Windows Subsystem for Linux and PyTorch.
- Early independent tests highlight strong local inference and gaming capabilities with DLSS 5, though sustained heavy AI workloads rapidly deplete battery life.
On October 16, 2026, Nvidia and major personal computer manufacturers launched RTX Spark laptops, introducing a new generation of Windows on Arm systems engineered to run frontier artificial intelligence models and autonomous agents locally. According to the NVIDIA Newsroom, initial systems from Asus, Dell, HP, Lenovo, Microsoft, and MSI are available immediately starting at $2,599, while shattered.io reports that compact desktop configurations are scheduled to arrive in November 2026.
The systems combine Nvidia's Grace central processing unit and Blackwell RTX graphics processing unit into a single system-on-a-chip (SoC) equipped with up to 128GB of unified memory. By offering up to 1 petaflop of FP4 compute, RTX Spark aims to eliminate developer reliance on distributed cloud infrastructure for agent development, while giving Windows users an architecture comparable to high-memory Apple silicon.
Superchip Architecture and Hardware Specifications
At the core of the RTX Spark platform is the N1X processor, also known as the GB10 SoC. As detailed by PCMag, the chip is fabricated on TSMC's 3-nanometer process node and operates across a configurable thermal design power ranging between 45 watts and 80 watts. The processor links the Grace CPU and Blackwell GPU using Nvidia's NVLink-C2C bidirectional interconnect, providing 600 gigabytes per second of bandwidth to eliminate traditional PCI Express bus bottlenecks.
According to PCMag, the Grace CPU incorporates Arm Cortex-X925 performance cores and Arm Cortex-A725 efficiency cores, developed in collaboration with MediaTek, according to the NVIDIA Newsroom. The platform is offered in two silicon tiers: a flagship 20-core version (10 performance and 10 efficiency cores) paired with a 6,144 CUDA-core Blackwell GPU and 192 fifth-generation Tensor Cores, and an 18-core version with 5,120 CUDA cores.
The memory subsystem operates across a 256-bit, 16-channel LPDDR5X interface reaching speeds up to 9,400 Mbps and peak unified bandwidth of 300 GBps. As reported by shattered.io, this unified architecture allows both the CPU and GPU to address the same memory pool. On 128GB models, users can dynamically allocate up to 96GB directly to the GPU, overcoming the traditional 24GB VRAM ceiling found on mobile workstations.
In addition to the GPU and CPU, the chip integrates a dedicated neural processing unit producing 40 trillion operations per second (TOPS), fulfilling Microsoft's certification requirements for Copilot+ features including Live Captions and Windows Studio Effects.
Local AI Agents and the Windows Software Ecosystem
Nvidia is positioning RTX Spark as a single machine capable of coding, testing, and deploying local software agents. The company claims that compared to Intel's Panther Lake 12Xe, RTX Spark delivers 4.6 times the performance on coding agents, 2.7 times on model fine-tuning, and 1.8 times on code compilation, according to the NVIDIA Technical Blog.
To establish an out-of-the-box development environment on Windows on Arm, Nvidia and Microsoft introduced several foundational toolsets. These include the first official PyTorch-CUDA build for Windows on Arm, native Python 3.15 integration, more than 2,000 precompiled Conda packages, and compatibility across 97 percent of the top 15,000 Python packages. Developers can work natively in Windows using Windows Developer Config or via Windows Subsystem for Linux (WSL).
For on-device intelligence, Nvidia released Nemotron 3.5 Bolt 74B-A9B, an open-source mixture-of-experts model featuring 74 billion total parameters and 9 billion active parameters. Using 2-bit quantization, the model fits within a 24GB footprint and is compatible with runtimes including llama.cpp, vLLM, Ollama, Unsloth, and LM Studio. Analysis by Thurrott.com notes that top-tier 128GB configurations can run models of up to 120 billion parameters with context windows reaching one million tokens.
Supported coding agents include GitHub Copilot, Hermes from Nous Research, OpenClaw, Perplexity, Pi, and JetBrains Junie Local. To manage agent operations safely, the platform incorporates sandboxing through Nvidia OpenShell and Microsoft Execution Containers (MXC), restricting unauthorized file modifications while letting local agents execute tasks in the background.
Performance, Battery Life, and Creative Workflows
Beyond artificial intelligence development, RTX Spark functions as a high-end platform for content creation and gaming. In creative workflows, Adobe has introduced beta releases of Photoshop and Premiere featuring native RTX Spark support accelerated by Nvidia TensorRT, enabling 8K 4:2:2 video scrubbing and 3D scene rendering up to 90GB, according to the NVIDIA Newsroom.
In independent evaluations conducted by The Verge, reviewers found that the platform handles gaming workloads reliably via Microsoft's Prism x86 emulation layer alongside native features like DLSS 5. In testing with an Asus ProArt P16 and HP OmniBook Ultra, Cyberpunk 2077 averaged 41 to 46 frames per second on Ultra settings at 2560x1600 resolution, climbing to 92 to 103 fps when DLSS Frame Generation was set to 4x. In Black Myth: Wukong, frame rates reached 70 fps at native 3270x2180 resolution on medium settings with DLSS Performance enabled.
Battery endurance, however, varied substantially based on workload. Reviewers testing systems equipped with 99.9 watt-hour batteries recorded approximately 5 to 5.5 hours of standard browser and document work at 50 percent screen brightness. Sustained local AI tasks, such as generating automated news briefings with Hermes and Qwen models, caused rapid battery depletion, fan noise, and elevated thermal output, demonstrating that intense inference remains power-heavy.
Pricing, Configurations, and Market Outlook
RTX Spark laptops occupy the premium tier of the portable PC market. According to The Verge, base configurations powered by the 18-core N1X chip start at $2,599.99 for systems like the Microsoft Surface Laptop Ultra and Asus ProArt P14 with lower memory capacities. High-tier 20-core models with 128GB of unified RAM scale up to $5,900 for the Surface Laptop Ultra and $6,999.99 for the Asus ProArt P16.
While costly, reporting from Engadget highlights that high-end RTX Spark systems enter the market below the $7,000 price point commanded by a fully equipped 128GB M5 Max MacBook Pro, while shattered.io notes the advantage of day-one CUDA ecosystem compatibility.
Nvidia founder and CEO Jensen Huang stated during Computex that Nvidia's goal was to "reinvent the PC" and focus "100% on Windows," while shattered.io noted that the initiative targets the roughly 1.5 billion active Windows PCs globally. With compact desktops scheduled to follow in November 2026, RTX Spark provides software engineers and enterprise researchers with a dedicated local hardware foundation for private agentic AI.
Questions people ask
What is NVIDIA RTX Spark?
RTX Spark is a system-on-a-chip platform combining an Arm-based Nvidia Grace CPU, a Blackwell RTX GPU, and up to 128GB of unified LPDDR5X memory to run generative AI models and autonomous agents locally on Windows.
Which manufacturers are selling RTX Spark PCs?
Laptops are currently available from ASUS, Dell, HP, Lenovo, Microsoft, and MSI, with compact desktop form factors arriving in November 2026.
How much do RTX Spark laptops cost?
Retail pricing starts at $2,599 for base 18-core configurations and reaches up to $6,999.99 for top-tier systems equipped with 20 CPU cores and 128GB of unified memory.
Can RTX Spark run local AI models without cloud access?
Yes. With up to 128GB of unified memory and 1 petaflop of FP4 compute, systems can run local large language models and coding agents such as Hermes, OpenClaw, and Nemotron 3.5 Bolt directly on device.
Sources
- NVIDIA RTX Spark PCs Available Today — A New Beginning for Windows PCsNVIDIA Newsroom, official source
- Code, Test, and Deploy on One Machine with NVIDIA RTX SparkNVIDIA Technical Blog, Oct 16, 2026, official source
- NVIDIA's RTX Spark Is Good For Windows, Even If You Don't Need Local AIEngadget, Oct 15, 2026
- A week with RTX Spark: everything we’ve learned using Nvidia’s new laptopsThe Verge, Oct 16, 2026
- Nvidia RTX Spark: Everything You Need to Know Before Our First TestsPCMag, Oct 15, 2026
- RTX Spark: Local AI First Steps ⭐Thurrott.com, Oct 16, 2026
- NVIDIA RTX Spark Launch: 128GB Memory Rivals Appleshattered.io, Oct 16, 2026
How this story was made: the newsroom picked it up from Google Search, Google News and theverge.com, gathered the full text of the sources above, and drafted it with AI assistance. Every factual claim was then checked against those sources before publishing (26 claims checked). Illustrations marked as AI-generated are not photographs. Spotted an error? Tell us.
Published October 17, 2026 at 00:18 UTC
