AI BriefingNVIDIA Expands Local AI Agent Ecosystem Based on RTX PCs and DGX Spark
Key point
NVIDIA unveils RTX Spark and OpenShell, expanding a local AI agent ecosystem equipped with both security and performance.
Details
At GTC Taipei, NVIDIA unveiled the NVIDIA RTX Spark, a new class of PC optimized for running personal AI agents. With 1 petaflop of AI compute power and 128GB of unified memory, RTX Spark meets the high computing demands of on-device agents and is designed to serve as a user's 'teammate' rather than just a simple tool.
For professional-grade solutions, NVIDIA introduced the NVIDIA DGX Station for Windows, which brings data center-class GPUs and CPUs to the desktop environment. This allows professionals to secure powerful AI inference performance while retaining the manageability and security of Windows.
To strengthen security and privacy, NVIDIA also announced the NVIDIA OpenShell runtime, developed in collaboration with Microsoft. Built on Microsoft's new security primitives, OpenShell ensures that agents run safely under user control and provides functionality to identify personal information and protect data being sent to cloud models.
Key technical updates include:
- Expanded application of the NVIDIA NemoClaw blueprint across the lineup, from GeForce RTX to DGX
- A 2x improvement in inference performance via multi-token prediction (MTP) in llama.cpp and vLLM
- Architecture redesigns for Adobe's Photoshop and Premiere, and Blender's support for DLSS 4.5 Ray Reconstruction
- Feature optimizations for NVIDIA Broadcast 2.2 and Project G-Assist
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.