NVIDIA Unveils RTX Spark and DGX Systems to Power Local AI Agents on PCs and Workstations

Image: NVIDIA · Source

NVIDIA launched its RTX Spark laptops alongside DGX Spark and DGX Station systems, delivering powerful local AI computing for personal agents, developers, and enterprises with Windows and Linux support starting now and expanding through year-end.

At the recent GTC Berlin event, NVIDIA announced an extensive lineup of local AI hardware designed to accelerate the next wave of AI agents and workflows on personal and enterprise computing platforms. Central to these offerings is the newly launched NVIDIA RTX Spark, a Windows PC platform combining an NVIDIA Blackwell RTX GPU and Grace CPU. RTX Spark delivers up to 1 petaflop of AI compute and 128GB of unified memory, enabling creators, developers, and gamers to run advanced local AI models, render large 3D scenes, and enjoy high-frame-rate gaming within a portable form factor. Major OEMs including ASUS, Dell, HP, Lenovo, Microsoft, and MSI have released RTX Spark laptops starting October 16, with Acer expected soon.

NVIDIA also introduced DGX Spark systems targeted at developers building and scaling AI applications on Linux. The DGX Spark 64GB model became available October 23, supporting large language models like Nemotron 3.5 Bolt 74B and Qwen 3.8 27B. Multiple DGX Spark units can be connected via NVIDIA's ConnectX-7 networking and Sync Cluster Assistant, enabling clusters up to eight systems strong, scaling to 1TB of unified memory for running extensive and concurrent AI workloads. Software improvements include Wi-Fi Assistant, updated agent skills, and a forthcoming Model Launcher to simplify deployment.

For enterprise teams requiring higher AI throughput, the DGX Station workstation offers up to 748GB coherent memory and 20 petaflops of AI performance, supporting multiuser deployments of frontier models such as GLM 5.3. The DGX Station runs Linux today, with a Windows variant launching in December. The Windows version will enable new manufacturing and engineering applications from partners like Dassault Systèmes and Siemens, combined with enterprise integration tools for security and device management. Robotics workflows are also supported via open source reference implementations for training, simulation, and synthetic data generation.

NVIDIA unveiled Recursive Flint, a new 40-billion-parameter local coding model optimized for RTX Spark, delivering fast, agentic coding performance and available open source. Additional day-zero support was announced for several AI models and inference accelerations for llama.cpp and vLLM engines tailored to NVIDIA local hardware.

Personal AI agents are a key highlight, with applications like Underdog leveraging fully local, private AI on RTX Spark. These agents run entirely on-device to protect user data privacy. NVIDIA’s ecosystem tools, including OpenShell runtime and execution containers, enable secure and efficient control of local AI agents integrated with creative and productivity applications.

This comprehensive portfolio of local AI solutions underscores NVIDIA's strategy to empower creators, developers, and enterprises with on-premises AI compute that minimizes cloud dependency while optimizing performance, security, and user control. NVIDIA's partners and developers can access playbooks, software updates, and optimized models to start leveraging these platforms today.