Skip to content
Neural Daily

ToolsNews

The first NVIDIA RTX Spark laptops go on sale

The first NVIDIA RTX Spark laptops went on sale on October 16 from ASUS, Dell, HP, Lenovo, Microsoft and MSI, starting at $2,599. For developers, the key change is native PyTorch CUDA support on Windows on Arm for the first time.

By: Mikola Máté Patrik · · 2 min read

NVIDIA RTX Spark laptops shown open in several positions over a circuit-pattern floor
Image: NVIDIA · Press kit · Official press image from NVIDIA Newsroom · Source

What's inside

RTX Spark combines a Blackwell RTX GPU and a Grace CPU, co-designed with MediaTek, on a single superchip with up to 128GB of unified memory and, NVIDIA says, 1 petaflop of FP4 AI performance (NVIDIA). The large shared memory lets laptops run models locally that previously needed a server or desktop.

The lineup: ASUS ProArt P14 and P16, Dell XPS 16, HP OmniBook Ultra and OmniBook X 14, Lenovo Yoga Pro 9n, Microsoft Surface Laptop Ultra, and MSI Prestige N16 Flip AI+ and Prestige N16. Compact desktops follow in November.

For developers

NVIDIA's developer blog says it worked with Microsoft to expand the Windows on Arm developer stack (NVIDIA Developer):

  • the first PyTorch CUDA build for Windows on Arm;
  • Python 3.15, the first release that is native on Arm by default;
  • over 2,000 ready-to-use Conda packages and a native Conda installer;
  • Arm-compatible versions of more than 97% of the top 15,000 Python packages;
  • WSL, offering the same Linux environment used for cloud deployment.

In NVIDIA's own tests against the Intel Core Ultra X9 388H, RTX Spark is 4.6x faster on coding agents, 2.7x on fine-tuning and 1.8x on compilation. These are vendor figures.

Local agents

Guided installers set up agents including GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi and WorkBuddy. NVIDIA OpenShell and Microsoft Execution Containers limit agent permissions with OS-level sandboxing.

For local agents NVIDIA recommends Nemotron 3.5 Bolt 74B-A9B 24GB, a 74-billion-parameter MoE model with 9 billion active parameters, packed into 24GB with 2-bit quantisation. It runs on llama.cpp, vLLM, Ollama, Unsloth and LM Studio, and its weights, training data and recipes are open. With NVIDIA PAIR, an RTX Spark can serve inference to other Windows, Linux and macOS machines on a home network.

Who it's for

Developers who build, fine-tune or test models and agents and want to do it on one machine, instead of or alongside the cloud. Arm-based Windows PCs lacked native CUDA and Python until now, which makes this the biggest change. The price is high and the performance numbers come from NVIDIA, so independent reviews are worth waiting for.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles