Nvidia has released an open-source tool called PAIR that directs local artificial intelligence requests across multiple devices within a home network. This software functions as a virtual router positioned between standard applications like Ollama or LM Studio and the computers they utilise. It distributes parallel tasks to available machines instead of burdening a single graphics card. In a demonstration, a cluster of three devices completed a task involving five subagents in under nine minutes, whereas a single laptop required nearly eighteen minutes. The system automatically identifies compatible hardware, which includes GeForce RTX cards from the 20 series onwards, RTX Pro workstations, DGX Spark units, and Apple silicon chips starting with the M4 model. All traffic between the machines is secured using MTLS encryption. The beta version is currently available for Windows, macOS, and Linux operating systems.
This development highlights Nvidia’s strategy to bind open-source AI development more closely to its proprietary hardware ecosystem. By creating software that optimises performance across diverse local setups, the company encourages users to rely on Nvidia GPUs for inference workloads. This approach mirrors the logic behind its recent $12.9 billion acquisition of Hugging Face, aiming to consolidate control over the local AI stack. The move suggests a shift from cloud-dependent models to distributed home computing that still depends heavily on Nvidia infrastructure.
- PAIR sits between existing tools and network devices
- Supported hardware ranges from RTX 20 series to Apple M4
- Beta software is available for Windows, macOS, and Linux
