Nvidia Turns Home Network into AI Cluster with PAIR

Nvidia launches PAIR, an open-source software router that distributes local AI requests among machines within the same household. An internal demonstration highlights significant time savings, and a beta version is available on major systems.

Hardware Strategy and Acquisition of Hugging Face

PAIR fits into Nvidia's strategy to bring open AI closer to its hardware. The tool supports GeForce RTX cards starting from the 20 series, RTX Pro workstations, DGX Spark, as well as Apple chips starting from the M4. This focus is also reflected in the acquisition of Hugging Face for $12.9 billion.

A Transparent and Secure Software Router

PAIR acts as a virtual router connecting tools like Ollama or LM Studio to computers on the home network, without requiring modifications to agents or applications. Instead of relying on a single GPU, it distributes requests across all available machines and aggregates the results for the requesting application. The tool automatically identifies compatible devices and secures traffic between them using MTLS encryption. PAIR is offered as open-source.

Measured Performance and Availability

During a demonstration, a task involving five sub-agents was completed by a cluster of three devices in just under 9 minutes, compared to 18 minutes on a single laptop. The beta version of PAIR can be used on Windows, macOS, and Linux. The goal is to reduce wait times by automatically distributing local AI workloads across all machines in the household.