Nvidia has launched the RTX Spark series of compact desktop PCs, featuring the 20-core Nvidia Grace CPU and a Blackwell-architecture GPU with 6144 CUDA cores. These Windows on Arm systems support up to 128 GB of unified memory and deliver AI performance of up to 1 petaflop. The devices include a 10 Gbps network interface for high-speed data transfer.
The RTX Spark systems are designed for local AI tasks, including running large language models with up to 120 billion parameters, real-time 3D rendering of scenes over 90 GB, and video editing. Developers can utilize CUDA and Linux tools via the Windows Subsystem for Linux (WSL), alongside the Hermes Agent platform and Unsloth integration for model fine-tuning.
New Nvidia Switchyard technology enables AI agents to distribute workloads between local and cloud resources. Additionally, Adobe software updates bring TensorRT-RTX acceleration to Premiere and Photoshop, while Illustrator and InDesign now feature native Arm support. Blender 5.3 has also introduced DLSS 4.5 Ray Reconstruction for the Cycles viewport.
The initial RTX Spark lineup includes models from Asus, Dell, and MSI, such as the Asus ProArt GR1X and MSI EdgeMesa N AI+. These systems are available starting November 18, 2026, with prices beginning at $4399. Future releases from Acer, Gigabyte, HP, and Microsoft are expected to follow.





