Nvidia announced the general availability of new RTX Spark compact desktops manufactured by ASUS, Dell, and MSI, with additional systems coming soon from Acer, GIGABYTE, HP, and Microsoft. Designed for local model fine-tuning, image and video generation, and autonomous agent execution, the dedicated Windows PCs feature a 20-core Nvidia Grace CPU alongside an Nvidia Blackwell RTX GPU. The machines deliver up to 128GB of unified memory and 1 petaflop of local AI performance.

The desktop hardware is paired with native Windows on Arm software ecosystem expansions, including integrations across Adobe applications, Blender, and local workflows in ComfyUI. The devices support continuous local execution of tasks via Hermes Agent or Nvidia NemoClaw blueprints, while utilizing Nvidia Switchyard to automatically route heavier reasoning tasks to cloud models when necessary.

By moving execution to dedicated local hardware, developers and enterprises can run persistent agents without incurring per-token cloud API costs for basic operations. The platform also addresses data security requirements by retaining processing tasks, code, and confidential creative assets directly on the local machine.

Why it matters

  • Enables engineers and creators to reduce cloud API burn by shifting continuous model workloads and agents to local hardware.

  • Provides dedicated local environments for secure handling of proprietary code, media, and enterprise data.

  • Accelerates developer adoption of hybrid AI architectures by automatically routing complex tasks between edge systems and cloud providers.

Source: blogs.nvidia.com