--- title: NVIDIA PAIR routes AI inference across idle home PCs,... description: At IFA 2026, NVIDIA unveiled PAIR (Personal AI Router), a free open-source tool that distributes AI inference requests across idle Macs and PCs on a local... canonical: https://daily.dev/posts/nvidia-pair-routes-ai-inference-across-idle-home-pcs-plus-other-local-ai-updates-from-ifa-2026-oemzmniav twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: NVIDIA PAIR routes AI inference across idle home PCs, plus other local AI updates from IFA 2026 | daily.dev og:description: At IFA 2026, NVIDIA unveiled PAIR (Personal AI Router), a free open-source tool that distributes AI inference requests across idle Macs and PCs on a local... og:url: https://daily.dev/posts/nvidia-pair-routes-ai-inference-across-idle-home-pcs-plus-other-local-ai-updates-from-ifa-2026-oemzmniav og:image: https://api.daily.dev/og/posts/OemZmniAv.png og:image:alt: NVIDIA PAIR routes AI inference across idle home PCs, plus other local AI updates from IFA 2026 og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # NVIDIA PAIR routes AI inference across idle home PCs, plus other local AI updates from IFA 2026 **[Collections](https://daily.dev/sources/collections)** · 3 min read · 0 upvotes · 0 comments ## Summary At IFA 2026, NVIDIA unveiled PAIR (Personal AI Router), a free open-source tool that distributes AI inference requests across idle Macs and PCs on a local network, relying on existing Ollama or LM Studio installs rather than a new inference engine. It routes whole requests to one node at a time and supports RTX 20-series+, Apple M4+, or DGX Spark hardware; NVIDIA's own benchmark showed two RTX 5090 machines completing a five-subagent workflow about 1.6x faster than one. Alongside PAIR, NVIDIA announced llama.cpp/vLLM optimizations giving up to 1.9x faster RTX inference, new RTX Spark Windows PCs from Lenovo and Acer launching in October, several new open-weight models tuned for local NVIDIA hardware, DLSS 4.5, and setup improvements for Hermes Agent, OpenClaw, and Perplexity Portable Computer, plus new local-inference benchmarking tools. ## Content ## What NVIDIA announced at IFA 2026 NVIDIA used IFA 2026 to push local AI harder, with a mix of new software tools, inference optimizations, and hardware. The most interesting piece is PAIR. ## NVIDIA PAIR: distributed inference across your home network PAIR (Personal AI Router) is a free, open-source tool that routes AI inference requests to idle Macs and PCs sitting on your local network. The idea is straightforward: if you're running an agentic workflow with multiple subagents firing off requests, PAIR distributes those requests across whichever machines are available rather than queuing everything on one device. A few things worth knowing about how it actually works: - It's not a new inference engine. It relies on Ollama or LM Studio already installed on each machine. - It routes complete inference requests to one node at a time — it doesn't split a single request across multiple GPUs. - Supported hardware: NVIDIA RTX 20-series or newer on Windows/Linux, Apple M4 or newer on macOS, or DGX Spark. In NVIDIA's own benchmark, two RTX 5090 PCs running Qwen3.6 35B A3B completed a five-subagent workflow about 1.6x faster than a single machine. That's a real speedup, though the conditions are specific enough that your mileage will vary. PAIR is currently in beta and can also help install Ollama or LM Studio and pull down models on paired machines. ## Faster inference on existing hardware Separately, new optimizations to llama.cpp and vLLM deliver up to 1.9x faster inference on RTX hardware. These are available through LM Studio and Ollama, so if you're already using either, you may see improvements without changing anything. ## New hardware: RTX Spark Windows PCs Lenovo and Acer are both releasing RTX Spark Windows PCs in October. Game support at launch comes from EA, Embark, and Ubisoft. ## New models optimized for local NVIDIA hardware Several new open-weight models were announced or highlighted alongside the hardware news: - Nemotron 3.5 Lightning - GLM-5.3-Flash - Qwen3.8-Flash-Next - LTX 2.5 - MiniMax-H3 - Muse Glimmer - DeepSeek v4 Flash ## Other updates - **Hermes Agent** now has simplified local setup for running agents directly on RTX and DGX hardware - **OpenClaw** and **Perplexity Portable Computer** also got local AI setup improvements - **DLSS 4.5** was announced - **DeepSeek Harness** and **MLPerf Client v2.0** add new benchmarking options for local inference The throughline across all of this is NVIDIA pushing inference off the cloud and onto hardware people already own. PAIR is the most novel piece — routing idle machines as inference nodes is a genuinely useful idea for anyone running multi-agent workflows at home, assuming you have compatible hardware sitting around. ## Questions this post answers ### What is NVIDIA PAIR and how does it distribute AI inference across home devices? PAIR (Personal AI Router) is a free, open-source tool from NVIDIA that routes AI inference requests to idle Macs and PCs on a local network. It relies on Ollama or LM Studio already installed on each machine, sends whole inference requests to one node at a time rather than splitting a request across GPUs, and supports RTX 20-series or newer, Apple M4 or newer, or DGX Spark hardware. It is currently in beta. _daily.dev surfaces releases like this for developers experimenting with distributed local inference setups._ ### How much faster is multi-machine inference with NVIDIA PAIR compared to a single PC? In NVIDIA's own benchmark, two RTX 5090 PCs running Qwen3.6 35B A3B completed a five-subagent workflow about 1.6x faster than a single machine. The speedup is workload-specific since PAIR routes complete requests to one node at a time rather than splitting a single request across GPUs, so results vary by hardware and task. _Developers benchmarking multi-agent workflows can track speedup claims like this via daily.dev._ ### What speedup do the new llama.cpp and vLLM optimizations provide on RTX hardware? New optimizations to llama.cpp and vLLM deliver up to 1.9x faster inference on RTX hardware. These improvements are available through LM Studio and Ollama, meaning users already running either tool may see performance gains without changing their setup. _Anyone tuning local inference speed can follow RTX and Ollama performance updates on daily.dev._ ## Similar posts on daily.dev - [I connected two PCs to one AI endpoint with Nvidia's new router, and got it serving an engine it doesn't support](https://daily.dev/posts/i-connected-two-pcs-to-one-ai-endpoint-with-nvidia-s-new-router-and-got-it-serving-an-engine-it-doe-uivsgfz7t) · XDA Developers · 0 upvotes · 0 comments - [NVIDIA Levels Up Local AI Agents Across RTX PCs and DGX Spark](https://daily.dev/posts/nvidia-levels-up-local-ai-agents-across-rtx-pcs-and-dgx-spark-du5cveehq) · NVIDIA · 0 upvotes · 0 comments --- Tags: [#nvidia](https://daily.dev/tags/nvidia), [#gpu](https://daily.dev/tags/gpu), [#distributed-systems](https://daily.dev/tags/distributed-systems), [#ollama](https://daily.dev/tags/ollama) [View this post on daily.dev](https://daily.dev/posts/nvidia-pair-routes-ai-inference-across-idle-home-pcs-plus-other-local-ai-updates-from-ifa-2026-oemzmniav) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"NVIDIA PAIR routes AI inference across idle home PCs, plus other local AI updates from IFA 2026","url":"https://daily.dev/posts/nvidia-pair-routes-ai-inference-across-idle-home-pcs-plus-other-local-ai-updates-from-ifa-2026-oemzmniav","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/nvidia-pair-routes-ai-inference-across-idle-home-pcs-plus-other-local-ai-updates-from-ifa-2026-oemzmniav"},"datePublished":"2026-09-03T18:02:34.815Z","dateModified":"2026-09-03T18:03:22.896Z","description":"At IFA 2026, NVIDIA unveiled PAIR (Personal AI Router), a free open-source tool that distributes AI inference requests across idle Macs and PCs on a local...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/f428fba0ff9fa7491ba7d7d363e0a7b1?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/f428fba0ff9fa7491ba7d7d363e0a7b1?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/nvidia-pair-routes-ai-inference-across-idle-home-pcs-plus-other-local-ai-updates-from-ifa-2026-oemzmniav","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"nvidia,gpu,distributed-systems,ollama","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"NVIDIA PAIR routes AI inference across idle home PCs, plus other local AI updates from IFA 2026"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/nvidia-pair-routes-ai-inference-across-idle-home-pcs-plus-other-local-ai-updates-from-ifa-2026-oemzmniav#faq","mainEntity":[{"@type":"Question","name":"What is NVIDIA PAIR and how does it distribute AI inference across home devices?","acceptedAnswer":{"@type":"Answer","text":"PAIR (Personal AI Router) is a free, open-source tool from NVIDIA that routes AI inference requests to idle Macs and PCs on a local network. It relies on Ollama or LM Studio already installed on each machine, sends whole inference requests to one node at a time rather than splitting a request across GPUs, and supports RTX 20-series or newer, Apple M4 or newer, or DGX Spark hardware. It is currently in beta. daily.dev surfaces releases like this for developers experimenting with distributed local inference setups."}},{"@type":"Question","name":"How much faster is multi-machine inference with NVIDIA PAIR compared to a single PC?","acceptedAnswer":{"@type":"Answer","text":"In NVIDIA's own benchmark, two RTX 5090 PCs running Qwen3.6 35B A3B completed a five-subagent workflow about 1.6x faster than a single machine. The speedup is workload-specific since PAIR routes complete requests to one node at a time rather than splitting a single request across GPUs, so results vary by hardware and task. Developers benchmarking multi-agent workflows can track speedup claims like this via daily.dev."}},{"@type":"Question","name":"What speedup do the new llama.cpp and vLLM optimizations provide on RTX hardware?","acceptedAnswer":{"@type":"Answer","text":"New optimizations to llama.cpp and vLLM deliver up to 1.9x faster inference on RTX hardware. These improvements are available through LM Studio and Ollama, meaning users already running either tool may see performance gains without changing their setup. Anyone tuning local inference speed can follow RTX and Ollama performance updates on daily.dev."}}]} ```