NVIDIA Launches RTX Spark Personal AI Supercomputing Platform

At GTC Berlin, NVIDIA introduced its full lineup of local AI hardware: RTX Spark for Windows PCs, DGX Spark for Linux developers, and DGX Station for enterprise Linux workloads, with a Windows version slated for December. The announcement also revealed new models, software updates, and open‑source workflows, including the Underdog Desktop Agent and the Recursive Flint coding model. These releases aim to enable personal agents, accelerated content creation, and scalable AI development on local machines.

NVIDIA introduced Nemotron 3.5 Super VL, a 120-billion-parameter open model delivering frontier accuracy for multimodal agentic tasks at one-fifth the size of largest reasoning models and up to 5x faster throughput. Two new edge models, Bolt 74B-A9B 24GB and Bolt E4B 4GB, were also released to optimize intelligence and speed for local agentic workloads and fast conversational agents on edge devices.

NVIDIA's RTX Spark integrates gaming, creative, and AI supercomputer functionalities into a single chip, offering up to 128GB unified memory and 1 petaflop AI compute. The platform debuts on October 16 with new software features and optimizations, including updates to GitHub Copilot and Adobe apps, and support for hundreds of games and anti-cheat tools.

Nvidia launched Nemotron 3.5 Super VL, an open 120 billion parameter multimodal model designed for continuous enterprise AI agents, alongside two compact Bolt models: Nemotron 3.5 Bolt 74B-A9B 24GB for local machines and Bolt E4B 4GB for fast voice assistants. Super VL uses a Mixture-of-Experts architecture, with only 12 billion parameters active per token, and Nvidia claims it matches the accuracy of Nemotron 3 Ultra on agent tasks with only about one-fifth the parameters, while offering up to five times the throughput on a single GPU.