NVIDIA Unveils PAIR Beta for Local AI Tasks at IFA

NVIDIA PAIR beta is a new free, open-source tool designed to route local AI inference tasks among compatible computers on the same local network. NVIDIA announced it on September 3 at IFA 2026, along with new optimizations for local models and confirmation that Windows devices with the RTX Spark platform from Lenovo and Acer will arrive in October.

NVIDIA PAIR beta for local networks

The name PAIR stands for Personal AI Router. According to NVIDIA, the tool is intended to distribute AI-agent inference tasks among available devices without sending processing to a cloud service. The goal is to use the computing power of multiple computers at home or in the office, including devices that would otherwise remain idle.

The software is currently in beta, and NVIDIA offers it for Windows, macOS, and Linux. Selected NVIDIA graphics processors, DGX Spark systems, and devices with an Apple M4 chip or newer are expected to be supported. The company did not provide a complete list of compatible hardware or a release date for the stable version.

Local processing can reduce the need to send sensitive data to external AI services. However, the tool’s availability alone does not automatically guarantee private deployment: permissions granted to agents, network configuration, and whether the user allows data to be moved to the cloud are also relevant. Since this is beta software, its practical operation and reliability across different network configurations have not yet been independently verified.

Higher throughput in llama.cpp according to NVIDIA

NVIDIA also announced optimizations for the llama.cpp project, which is used to run language models locally. According to the company’s data, they can increase throughput by up to 1.9× on a GeForce RTX 5090 graphics card. The optimizations are also expected to be available through the LM Studio and Ollama applications.

The “up to 1.9×” figure is based on NVIDIA measurements for a specific configuration and workloads. The available materials do not include independent comparative tests confirming this result in typical agent use or with different models and hardware. The resulting benefit may therefore vary between systems.

Windows PCs with RTX Spark are expected in October

NVIDIA confirmed that Lenovo and Acer plan to launch Windows devices built on RTX Spark in October 2026. The company introduced the platform with Microsoft on May 31, 2026. Microsoft said at the time that Windows for RTX Spark would receive optimizations for shared-memory management, task scheduling, and Prism emulation for x86 applications.

The announcement connects hardware focused on local AI with software intended to distribute tasks among multiple devices. RTX Spark systems from partners will also serve as a concrete test of interest in more powerful Windows computers designed for local AI workloads.

However, prices, exact sales dates, hardware configurations, and regional availability for specific Lenovo and Acer computers have not yet been fully specified. Independent tests of performance, power consumption, Windows on Arm compatibility, and the reliability of agent applications also remain outstanding.

What comes next

• NVIDIA will need to clarify the stable PAIR release date and the conditions for secure operation on a local network.
• Before the October launch, Lenovo and Acer are expected to publish prices and complete configurations for RTX Spark systems.
• Independent measurements across different models, applications, and hardware configurations will be important for assessing the claimed benefit of the optimizations.

Verified and updated: 09/04/2026 10:53