Tech

Nvidia's RTX Spark laptops go on sale, putting a Grace CPU and Blackwell GPU in Windows PCs

The first Windows laptops built on Nvidia's RTX Spark superchip went on sale globally on 16 October from ASUS, Dell, HP, Lenovo, Microsoft and MSI, starting at $2,599 and pitched at running AI agents locally.

Abstract illustration of a processor die with circuit traces fanning out, in orange and red.
Abstract illustration of a processor die with circuit traces fanning out, in orange and red.

Laptops built on Nvidia’s RTX Spark superchip went on sale worldwide on 16 October, four and a half months after Nvidia and Microsoft unveiled the chip at Nvidia’s GTC Taipei event during Computex. RTX Spark combines a Blackwell-generation RTX GPU and an Nvidia Grace CPU in a single package, and Nvidia is pitching the resulting machines as Windows PCs designed to run personal AI agents locally, alongside creative work and gaming. Prices start at $2,599, according to Nvidia, with pricing and configurations varying by region.

The launch line-up

Nvidia says the laptops are available through manufacturers’ websites and major retailers. Nine models from six manufacturers are on sale at launch:

Manufacturer Models
ASUS ProArt P14, ProArt P16
Dell XPS 16
HP OmniBook Ultra, OmniBook X 14
Lenovo Yoga Pro 9n
Microsoft Surface Laptop Ultra
MSI Prestige N16 Flip AI+, Prestige N16

Nvidia says more systems will follow. When it announced the chip in May, it said Acer and GIGABYTE models would come later and that compact desktops were also planned. Engadget reported that Microsoft’s Surface Laptop Ultra, which opened for pre-orders on 7 October, starts at $2,599.

What is inside

RTX Spark is Nvidia’s own system-on-chip for Windows PCs, so Nvidia supplies both the processor and the graphics rather than a discrete GPU paired with another company’s CPU. The platform runs Windows on Arm; Nvidia’s launch materials highlight the first PyTorch-CUDA build for Windows on Arm, and Engadget describes the chip as Arm-based. Nvidia’s published specifications are:

Component Nvidia’s specification
GPU Blackwell RTX GPU with 6,144 CUDA cores and fifth-generation Tensor Cores supporting FP4
CPU 20-core Nvidia Grace CPU, with a custom design developed in collaboration with MediaTek
Link NVLink-C2C between CPU and GPU
Memory Up to 128GB of unified memory
AI performance Up to 1 petaflop (FP4), vendor-reported

Memory and other configuration details vary between models. Large pools of unified memory are central to the pitch: they let bigger AI models run on the laptop itself rather than in the cloud.

Built around agents

Most of Nvidia’s launch announcement is about software, and in particular about AI agents that carry out tasks on the PC:

  • Agent setup. Guided setup and installers for agents including GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi and WorkBuddy, intended to work without the command line.
  • Permissions. Nvidia’s OpenShell secure runtime and Microsoft Execution Containers are designed to help users control what agents are allowed to do. In May, Nvidia said OpenShell can also route queries to local models depending on a user’s privacy settings.
  • Local models. On supported configurations, Nvidia offers Nemotron 3.5 Bolt, a 74-billion-parameter model it says is optimised for tool calling and multistep tasks.
  • Home network inference. A feature called Nvidia PAIR lets an RTX Spark machine serve AI inference to compatible Windows, Linux and macOS computers on the same home network.
  • Developers. Alongside the PyTorch-CUDA build, Nvidia claims support for 97% of the top 15,000 Python packages.

On the creative and gaming side, Adobe’s Photoshop and Premiere offer native RTX Spark support through beta releases available at launch, and Nvidia says the laptops can play AAA games at 1440p at more than 100 frames per second. Those are Nvidia’s figures; independent testing will show how they hold up across games and models.

“RTX Spark brings the power of NVIDIA AI, RTX graphics and CUDA together with Microsoft Windows in one PC,” Nvidia’s founder and chief executive, Jensen Huang, said in the announcement. Yusuf Mehdi, an executive vice president at Microsoft, said: “We’re building Windows to be the platform for hybrid intelligence,” describing AI that runs locally when it makes sense and reaches the cloud when needed.

What it means

Analysis. RTX Spark brings together several things Nvidia already does well, including its GPUs, its CUDA software platform and its data-centre CPU design, into a single chip for consumer laptops. The emphasis at launch is not raw graphics but local AI: lots of memory, a local model tuned for agent work, and a permissions layer built with Microsoft.

The starting price places these machines at the premium end of the Windows laptop market, aimed at creators and developers who expect to make use of the extra memory and AI performance. The practical questions are the ones that apply to any new platform. How smoothly do everyday Windows applications run on Arm? How do battery life and sustained performance compare with existing premium laptops? And will the agent software become something buyers use every day? Developer tooling such as the native PyTorch-CUDA build addresses part of the first question; independent reviews and wider availability, including the planned desktops, will answer the rest.

Sources

See All

More on Nvidia

See All

Type to search. ↑ ↓ to navigate, ↵ to open.