Verify RTX Spark and DGX Station Support for Local AI Agents

NVIDIA is expanding its lineup of hardware designed to run artificial intelligence models locally on devices rather than in the cloud. The company showcased these updates at its GTC Berlin event, introducing new software features for its RTX Spark and DGX product lines. These changes allow developers and enterprise teams to build and run AI applications with greater capacity and manageability.

In plain English

Local AI refers to running artificial intelligence models directly on a computer or server instead of sending data to remote cloud servers. This approach keeps user data on the device, which can improve privacy and reduce latency. It requires hardware with significant memory and processing power to handle complex tasks like coding assistance or content creation without external network dependencies.

The background

RTX Spark launched last week and powers personal agents on Windows PCs. It offers up to 128 gigabytes of unified memory and one petaflop of AI compute. The platform supports accelerated content creation and high-performance gaming. Underdog, a desktop agent application, is now available for RTX Spark. It uses a model-specific inference engine to run models entirely on the device.

DGX Spark supports developers building and scaling applications on Linux. New 64GB systems start rolling out from partners on October 23. These systems can run models including Nemotron 3.5 Bolt 74B and Qwen3.8 27B. Developers can connect multiple DGX Spark systems via built-in networking to run advanced coding models.

What changes now

DGX Spark now supports clusters of up to eight units in manual configuration. This setup provides one terabyte of unified memory for frontier-class models. NVIDIA Sync Model Launcher will be available later this month to simplify access to local AI. It downloads models and containers suited to the configuration and connects them to OpenCode.

DGX Station serves enterprise teams running frontier workloads on Linux. A Windows version arrives in December with support for many manufacturing apps. The system improves manageability with seven new integrations. NVIDIA Cloud Functions is coming in early access to DGX Station. Three new open source reference workflows for robotics are also available.

Key terms explained

When evaluating local AI hardware, teams usually verify that the device meets the memory requirements for the specific models they intend to run. Check whether the operating system version is supported by the vendor’s software stack. Ensure that network configurations allow for secure local communication if clustering multiple devices. Review documentation to understand how data remains on-device during inference tasks.

What to do and how to stay safe: NVIDIA

• Verify that your RTX Spark or DGX hardware matches the system requirements for the AI models you plan to deploy.
• Check if your organization needs to update network settings to support local clustering or multi-device communication.
• Review vendor documentation to ensure data privacy policies align with running models entirely on local devices.
• Monitor vendor announcements for software updates that add support for new AI models or management features.

General security guidance from the Vector Update newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

When is the Windows version of DGX Station available?

The Windows version of DGX Station launches in December with support for many manufacturing apps and agent harnesses.

Can DGX Spark systems be clustered together?

Yes, DGX Spark now supports clusters of up to eight units in manual configuration for larger models and concurrent agents.

What new software is coming for DGX Spark this month?

NVIDIA Sync Model Launcher will be available later this month to simplify access to local AI on single or clustered systems.