[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-nvidia-pair-personal-ai-router":3,"topics-all":38,"news-related-107277b2-2c3f-490b-bb68-a1432e723649":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"107277b2-2c3f-490b-bb68-a1432e723649","英伟达开源 PAIR：把家里闲置显卡串成一座个人 AI 数据中心","英伟达发布开源工具 PAIR，可将家庭中英伟达 RTX 20 及更新显卡、RTX Pro、DGX Spark 以及苹果 M4 以上芯片的闲置算力汇集起来，用于本地 AI 推理和智能体工作负载，通过六位数配对码 + mTLS 加密，支持 Windows、Linux、macOS。","家里的游戏卡和工作站，平时只在跑游戏和偶尔训练时才醒着，大量时间都处在「亮屏空跑」的状态。英伟达本周发布的一款开源工具盯上的正是这部分沉睡的算力：Personal AI Router（简称 PAIR）允许用户把家中多台兼容设备串成一个本地 AI 计算池，统一用于本地推理和智能体工作负载。\n\n## PAIR 在做什么\n\nPAIR 的目标很直白——把分散在书房、客厅、卧室里的多台电脑组合起来，对外提供「一台更大电脑」的算力体验。它会持续监测每台设备是否闲置：当你回到主力机前开始玩游戏或者跑重负载任务，对应设备会自动退出 PAIR 池子，把 GPU 资源还给你；任务结束后重新加入。整套调度策略围绕「不打扰人」展开，不会出现你打着游戏帧率却掉一半的尴尬。\n\n为照顾 Apple Silicon 用户，PAIR 同时支持苹果 M4 及以上的芯片。系统兼容性方面，PAIR 已经覆盖 Windows、Linux 和 macOS 三大主流桌面环境。\n\n## 设备门槛与配对流程\n\n兼容设备列表覆盖范围相当广：英伟达 GeForce RTX 20 系列及更新显卡、英伟达 RTX Pro GPU、以及 DGX Spark 都在支持范围内。配对流程做了简化——用户只需要在每台设备上输入同一个六位数代码即可完成组网。这意味着不需要复杂的端口映射，也不需要公网 IP，家庭局域网内部就能完成整套握手。\n\n通信安全方面，PAIR 使用 mTLS（Mutual Transport Layer Security）建立双向可信的加密通道。设备之间互相验证身份，避免被同网段的恶意节点冒充接入，趁机拉走本地数据或注入恶意推理任务。对本地 AI 工作流而言，这种端到端的可验证链路是把家庭设备暴露给网络前必须解决的一道关。\n\n## 它改变的是什么\n\nPAIR 的最大意义不在于「多机推理」本身——这条路线在分布式训练领域早有成熟方案；它的意义在于把门槛拉到普通消费者级别。过去想跑一个本地大模型，要么买一张大显存的专业卡，要么租用云端算力。PAIR 给出了第三条路：把你已经买过的、平时空置的消费级 GPU 利用起来，几张 RTX 4090 串起来就能跑比单卡更大的模型，本地响应、低延迟、数据不外传。\n\n对智能体（Agent）场景尤其关键。当 Agent 跑长链路任务、需要多轮工具调用时，单卡显存往往是最先撞到的墙；多机协同直接把可用显存和并行度拉高，Agent 不用再为「这个上下文放不放得下」做妥协。\n\n## 「个人数据中心」叙事的边界\n\n英伟达这次没有把 PAIR 包装成云服务替代品——它强调的是本地优先（local-first）路线，所有推理和数据都留在用户自己的设备上。这条叙事在监管收紧、企业 IT 收紧自带设备（BYOD）策略的当下，恰好踩中了不少个人开发者和小型团队的痛点。\n\n但需要看到边界：PAIR 解决的是「同一屋檐下」的算力聚合，并不能突破单台设备的物理上限；六位数配对码在便利和安全之间做了取舍，真要商用还需要更严格的鉴权链路；另外，无线网络的带宽和延迟会成为瓶颈，PCIe 总线级别的协同它并不解决。所以它更像是一台家用 NAS 级别的「个人 AI 基础设施」，而不是一台真正的数据中心。\n\n无论如何，对那些家里堆着几张显卡、却苦于单卡跑不动本地大模型的人来说，PAIR 至少把「攒机当数据中心」的门槛，从一篇博客教程压到了输入六位数字。开源地址见英伟达官方页面（https:\u002F\u002Fwww.nvidia.com\u002Fen-us\u002Fai-on-rtx\u002Fpersonal-ai-router\u002F），社区已经在讨论把它和 Ollama、vLLM、ExLlamaV2 等本地推理框架接通的玩法。","https:\u002F\u002Fwww.theverge.com\u002Fai-artificial-intelligence\u002F989435\u002Fnvidia-pair-personal-ai-router-home-local-llm-compute-tool-rtx-macbook","474eef8c-e0c3-46cf-adee-c089558220f9",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"fca9258a-9430-455a-b95d-b9fae5e373a8","ai-inference",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"e0d31e94-ce47-4c8f-831c-d3d2926d42f3","hardware",{"id":19,"name":20,"slug":20,"description":14,"color":14},"8dac812d-3839-4abe-a855-5f56ec9515fd","nvidia",{"id":22,"name":23,"slug":23,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"2fd32d0d-40a4-4ea9-aa82-4ec5751d9de9","en","NVIDIA open-sources PAIR: turn idle home GPUs into a personal AI data center","NVIDIA has open-sourced PAIR (Personal AI Router), an aggregator that pools idle compute across NVIDIA RTX 20 and newer GPUs, RTX Pro, DGX Spark, and Apple M4+ chips on a home network, exposing it as a single resource for local AI inference and agentic workloads. Devices pair via a six-digit code and communicate over mTLS; Windows, Linux, and macOS are all supported.","Gaming GPUs and workstations spend most of their day idling — fans spinning, screens lit, silicon doing nothing. NVIDIA's newly open-sourced Personal AI Router (PAIR) targets exactly that idle capacity: it lets users stitch together compatible machines on a home network into a single local AI compute pool, shared across inference and agentic workloads.\n\n## What PAIR actually does\n\nPAIR is, at its core, a local compute aggregator. It continuously monitors whether each device in the pool is idle. When the user returns to a primary machine to game or kick off a heavy task, that device automatically leaves the PAIR pool and returns its GPU to local use; once it goes idle again, it rejoins. The whole scheduling policy is built around not getting in the user's way — there is no scenario where your game frame rate mysteriously halves because PAIR decided to borrow your GPU.\n\nFor Apple Silicon users, PAIR supports M4 and above. On the OS side, it already covers Windows, Linux, and macOS.\n\n## Hardware bar and pairing flow\n\nThe supported device list is generous. NVIDIA GeForce RTX 20-series and newer cards, NVIDIA RTX Pro GPUs, and DGX Spark are all in scope. Pairing has been simplified — users enter the same six-digit code on each device to complete network setup. There is no port forwarding, no public IP required; the entire handshake happens inside the local network.\n\nOn the security side, PAIR uses mTLS (Mutual Transport Layer Security) to establish mutually authenticated encrypted channels between devices. Each node verifies the others, so a rogue machine on the same LAN segment cannot impersonate a legitimate device, exfiltrate local data, or inject malicious inference jobs. For local AI workflows that touch proprietary prompts or private documents, this end-to-end verifiable link is a non-negotiable prerequisite.\n\n## Why this matters\n\nPAIR is not the first distributed inference framework — that space has long been solved at the data-center scale. What makes PAIR notable is that it pulls the threshold down to consumer hardware. Running a local LLM used to mean either buying a single high-VRAM professional card or renting cloud GPUs. PAIR offers a third path: aggregate the consumer cards you already own, and several RTX 4090s in the same house can host a model that no single card could fit.\n\nFor agent workloads the impact is sharper. Agents running long tool-call chains typically hit the VRAM wall first; multi-device pooling directly expands both available memory and parallelism, so agents no longer have to make context-fit compromises on every step.\n\n## Where the \"personal data center\" framing stops\n\nNVIDIA is not positioning PAIR as a cloud replacement. The pitch is local-first: all inference and data stay on the user's own devices. That framing lands well in a moment of tightening AI regulation and stricter enterprise BYOD policies, where individual developers and small teams need ways to keep sensitive workloads on-prem.\n\nBut the limits are real. PAIR only aggregates compute within one physical network; it does not lift the per-device ceiling. The six-digit pairing code trades security for convenience — any real production deployment will need stronger authentication. And wireless LAN bandwidth and latency become the binding constraint, since PAIR does not solve PCIe-bus-level coordination. So think of it as a NAS-class \"personal AI infrastructure,\" not a true data center.\n\nStill, for anyone with a small stack of GPUs at home who has been unable to fit a local LLM into a single card, PAIR at least lowers the bar from \"follow a tutorial\" to \"type six digits.\" The project page is on NVIDIA's RTX site (https:\u002F\u002Fwww.nvidia.com\u002Fen-us\u002Fai-on-rtx\u002Fpersonal-ai-router\u002F); the community is already discussing wiring it up with Ollama, vLLM, and ExLlamaV2.","nvidia-pair-personal-ai-router","2026-09-05T06:25:00Z","2026-09-05T03:03:52.592213Z","2026-09-05T03:03:52.592221Z",true,"agent",196,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"61cf8d85-c751-4da2-9aae-10b645415ec9","英伟达发布开源工具 PAIR,把家里电脑连成个人 AI 推理集群","nvidia-pair-personal-ai-router-local-inference","2026-09-09T02:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"42b7939c-1b44-43b8-95cf-a8fc2204560d","NVIDIA 开源 Personal AI Router，把家里 RTX 与 Mac 拼成本地 AI 集群","nvidia-personal-ai-router-pair-beta","2026-09-04T03:20:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"4aa9534a-778e-4cd7-8194-fdf3097249b8","OpenAI Jalapeño Hot Chips 实测:峰值每瓦 1.9×,延迟压到 1 秒","openai-jalapeno-hot-chips-benchmark-2026","2026-08-26T02:00:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"a235e2ab-5b61-47b2-a85a-5f8a1d438624","AMD 收购 Taalas:把\"为单一模型造芯\"的路子,搬进 Instinct 体系","amd-acquires-taalas-hardwired-inference-silicon","2026-08-09T06:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"30a147aa-e3ed-475c-b15f-9e5ffce6ffc9","英伟达 Vera CPU：DeepInfra 实测 Agent 编排提速 2.2 倍","nvidia-vera-cpu-agent-orchestration","2026-07-22T04:50:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"f4aad332-1f97-4cb2-96a4-37d8d2980728","英伟达BW首秀RTX Spark：笔记本本地跑120B大模型","nvidia-rtx-spark-120b-laptop","2026-07-12T12:00:00+00:00"]