Particle.news
Download on the App Store

RTX Spark Laptops Arrive With Unified Memory and Native CUDA for Developers

The new Arm-based machines promise larger local AI models on a single device while price, battery life, and app compatibility will shape who actually adopts them.

Overview

  • Major OEMs and Microsoft began shipping the first RTX Spark notebooks in mid-October, led by Microsoft’s Surface Laptop Ultra built on NVIDIA’s RTX Spark system-on-chip and Blackwell GPU.
  • NVIDIA says the multi-year project delivered a 1‑petaflop Blackwell GPU and shared unified memory that on 128GB systems allows roughly 96GB to be allocated to GPU tasks, enabling larger models to run locally.
  • Early hands-on reviews praise the GPU performance and unified‑memory architecture for creator and developer workloads while flagging trade-offs in battery life, thermals, and real-world CPU performance.
  • Microsoft and NVIDIA are promoting native CUDA support on Windows on Arm, including promises of CUDA-accelerated PyTorch, a move that has already drawn interest from developers who depend on NVIDIA’s software stack.
  • High price points and remaining Windows on Arm app‑compatibility gaps are the main barriers to broader adoption, with top Surface Ultra configurations listed at $5,899.99 and entry options starting around $2,599.