Following this morning’s launch of the Nemotron 3.5 Lightning model, the NVIDIA Local AI blog has announced a series of major updates aimed at developers and enthusiasts building and running AI models locally. These updates bring substantial enhancements to the DGX Spark ecosystem, alongside performance boosts for open models running on consumer hardware.
Scaling Up with DGX Spark
For developers leveraging NVIDIA’s DGX Spark, two major updates are streamlining cluster management and expanding functionality:
- NVIDIA Sync Cluster Assistant: Users can now connect two or more DGX Spark systems into a high-speed cluster. This assistant streamlines networking, workload routing, and automated health monitoring.
- Native Chrome & Resource Monitoring: Google Chrome is officially arriving as a native ARM64 Linux build, helping users get more out of their DGX Spark. Furthermore, a new NVIDIA Sync Resource Monitor will provide real-time and historical views of CPU and GPU usage, whether you are monitoring a single system or an entire cluster.
High-Speed Local AI on RTX Hardware
The push for local AI continues to benefit consumer hardware, particularly for those building ultra-enthusiast PCs.
- Meta’s Muse Glimmer: This 30B open-weight model has been optimized for always-on local agents. As we’ve noted in our previous coverage of the GeForce RTX 5090, the flagship card provides massive compute power and can now run Muse Glimmer locally at an impressive rate of more than 200 tokens per second.
- LTX-2.5 Video Generation: Creating high-quality video locally is getting a major upgrade with LTX-2.5. The open video model now supports multishot generation and generative editing. Thanks to optimizations for NVIDIA RTX GPUs, DGX Spark, and DGX Station, users will see a 2x performance boost alongside a 40% reduction in memory usage.
More Open Models Accelerated by NVIDIA
The local AI ecosystem continues to grow, and NVIDIA is accelerating a wide range of new open models for developers to explore. The expanded lineup includes:
- Cosmos 3 Edge
- MiniMax-H3
- Poolside’s Laguna S 2.1
- DeepSeek-V4-Flash
- Thinking Machines Lab’s Inkling-Small
- Unsloth Desktop
- Wan-Animate-2
Related NVIDIA Links: