Bringing the Spark to Europe: Local Intelligence Powers Next Wave of Agents

At NVIDIA's GTC Berlin event, the company unveiled its full range of local AI hardware. RTX Spark, launched last week, enables personal agents on Windows PCs, while DGX Spark supports developers in building and scaling applications on Linux, and DGX Station caters to enterprise teams running high-performance workloads on Linux, with a Windows version set to launch in December.

Some of the key announcements include the availability of Underdog Desktop Agent for RTX Spark, DGX Spark 64GB model available on October 23, and enhancements to NVIDIA Sync to support eight DGX Sparks and introduce a Model Launcher. The October DGX Spark software update adds Wi-Fi Assistant and updates DGX Spark Skills, while a new Windows version of DGX Station is slated to launch in December, with support for various manufacturing apps and agent harnesses.

Three new open-source reference workflows for robotics on DGX Station were introduced, along with NVIDIA Cloud Functions set to roll out in early access to DGX Station. A new local coding model, Recursive Flint, is now accessible for NVIDIA local hardware. Day-0 support for numerous AI models like Nemotron, GLM, DeepSeek, Qwen, MiniMax, and FLUX.3 has also been added.

Focusing on software, NVIDIA has also announced new accelerations for llama.cpp and vLLM inference engines and a collection of checkpoints optimized for local AI hardware. Attendees at GTC Berlin can learn more about these local AI announcements and explore demos and sessions.