Itmedia iconItmediaSep 7, 2026 ~3 min source read

NVIDIA to Ship RTX Spark Windows PCs in October; PAIR beta lets LAN PCs share local AI work

NVIDIA showed RTX Spark–powered Windows systems at IFA and published a beta of NVIDIA PAIR (Personal AI Router), software that spreads inference tasks across GPUs on the local network to accelerate agentic and LLM workloads.

Share this story

Send the public story page.

Useful takeaways from this story.

NVIDIA PAIR beta pools GPUs across a LAN and dynamically routes AI agent inference to available machines to speed local AI tasks.

PAIR is offered for Windows, Linux and macOS and can work with common local LLM runtimes such as Ollama and LM Studio.

NVIDIA confirmed that the first Windows PCs built around its RTX Spark system-on-chip will reach customers in October. At IFA 2026 the company highlighted partner systems — Lenovo's Yoga 9n 2-in-1 and an Acer small-form-factor RTX Spark desktop among them — and described software steps intended to make local AI more practical on consumer and prosumer hardware.

What RTX Spark systems aim to change

NVIDIA PAIR (Personal AI Router) beta

NVIDIA published a beta of PAIR, short for Personal AI Router. PAIR is client/server software that discovers GPU-equipped machines on the same LAN and distributes AI inference tasks across them. The software's behavior is to split agent workloads and subagents and dispatch those subtasks to the best available devices on the network, which can reduce latency and increase throughput for local AI tasks.

  • Platform support: PAIR is available for Windows, Linux and macOS. That allows mixed networks (for example, a Mac and several Windows gaming PCs) to participate.
  • Runtime compatibility: NVIDIA mentions it working with local LLM runtimes such as Ollama and LM Studio, so users who run those local model servers can leverage PAIR to scale inference across machines.
  • Hardware scope: PAIR is intended to aggregate GeForce and NVIDIA RTX GPUs and can include NVIDIA's DGX Spark systems. NVIDIA says PAIR can also use other vendors' GPUs on the LAN, not strictly NVIDIA cards.

How this changes local AI workflows

PAIR lets a host machine offload parts of agentic or LLM inference to idle GPUs in other PCs on the same LAN. The result is lower latency and higher effective parallelism compared with running everything on one device. That makes it easier to run larger models or more concurrent agents locally without sending data to a cloud service. PAIR's cross-platform client means, for example, a Mac with an M-series chip can tap GPU cycles on an RTX-powered Windows desktop for faster inference.

  • Set up: Users will need to install PAIR on each participating machine and ensure network visibility between them. Wired connections are recommended for consistent performance.
  • Scope and scale: PAIR is a beta release. Its effectiveness will depend on the mix of hardware on the LAN, the models used, and the quality of the network.

More context around this story.

Loading more related stories...

Keep reading in the app

Open the app view to save this story, compare related coverage, and continue from the same source.

Open in app