Particle.news
Download on the App Store

Nvidia Unveils RTX Spark Superchip to Put Big AI Models on Windows PCs

Up to 128 GB of shared memory and NVLink between the CPU and GPU could let everyday laptops run large AI models locally.

Overview

  • Nvidia announced the RTX Spark superchip at GTC Taipei as a unified Arm-based CPU+GPU design that pairs a Blackwell RTX GPU with a 20-core Grace CPU connected by NVLink.
  • The chip is claimed to deliver up to 1 petaflop of AI performance and support as much as 128 GB of unified memory so the CPU and GPU can access the same data without copying it back and forth.
  • Nvidia named Microsoft, MediaTek and major PC makers including ASUS, Dell, HP, Lenovo, Microsoft Surface, MSI, Acer and GIGABYTE as partners and said devices as thin as 14 mm are planned to ship in fall 2026.
  • Coverage since the launch stresses open questions that will determine real-world impact: independent benchmarks, Windows-on-Arm app compatibility, thermal and battery behavior in thin designs, pricing, and manufacturing scale.
  • If RTX Spark meets its specs, users could run more AI workloads on their devices rather than in the cloud, a shift that would change latency, costs, and privacy trade-offs and likely spur competitive responses from Apple, Qualcomm, AMD, and Intel.