<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[Google Cloud Next ’26：TPU 8t 与 TPU 8i 双芯发布，训练与推理分工，算力效率较上代最高提升 3 倍]]></title><description><![CDATA[<p dir="auto">Google 在 Cloud Next ’26 大会上宣布推出第八代 TPU，分为两款独立架构：TPU 8t（专注训练）和 TPU 8i（专注推理），均将于今年晚些时候正式发布。<br />
TPU 8t：训练利器<br />
单个 TPU 8t 超级 Pod 可扩展至 9,600 颗芯片、2 PB 共享高带宽内存，提供 121 ExaFlops 算力，计算性能较上代提升近 3 倍。同时集成 10 倍更快的存储访问，并通过 Virgo 网络与 JAX/Pathways 软件支持最多百万芯片的近线性扩展。<br />
TPU 8i：推理引擎<br />
TPU 8i 搭载 288 GB 高带宽内存与 384 MB 片上 SRAM（较上代增加 3 倍），ICI 互连带宽提升至 19.2 Tb/s，新增片上集合加速引擎（CAE）将延迟降低最多 5 倍。整体性能每美元效率提升 80%，相同成本下可服务近两倍的用户量。<br />
能效与系统协同<br />
两款芯片均采用 Google 自研 Axion ARM CPU 主机，性能每瓦功耗较上代 Ironwood 提升最多 2 倍，并配备第四代液冷技术，支持原生 JAX、PyTorch、SGLang、vLLM 等主流框架。</p>
<p dir="auto"><a href="https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/eighth-generation-tpu-agentic-era/" target="_blank" rel="noopener noreferrer nofollow ugc">Google</a></p>
<p dir="auto"><div class="card col-md-9 col-lg-6 position-relative link-preview p-0">



<a href="https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/eighth-generation-tpu-agentic-era/" title="Our eighth generation TPUs: two chips for the agentic era">
<img src="https://storage.googleapis.com/gweb-uniblog-publish-prod/images/two_chips_for_the_agentic_era_hero.width-1300.png" class="card-img-top not-responsive" style="max-height: 15rem;" alt="Link Preview Image" onerror="this.parentElement.remove()" />
</a>



<div class="card-body">
<h5 class="card-title">
<a class="text-decoration-none" href="https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/eighth-generation-tpu-agentic-era/">
Our eighth generation TPUs: two chips for the agentic era
</a>
</h5>
<p class="card-text line-clamp-3">An overview of Google’s eighth generation TPUs, built for the agentic era.</p>
</div>
<a href="https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/eighth-generation-tpu-agentic-era/" class="card-footer text-body-secondary small d-flex gap-2 align-items-center lh-2">



<img src="https://blog.google/favicon.ico" alt="favicon" class="not-responsive overflow-hiddden" style="max-width: 21px; max-height: 21px;" onerror="this.remove()"/>





<p class="d-inline-block text-truncate mb-0">Google <span class="text-secondary">(blog.google)</span></p>
</a>
</div></p>
]]></description><link>https://welinux.com//topic/22/google-cloud-next-26-tpu-8t-与-tpu-8i-双芯发布-训练与推理分工-算力效率较上代最高提升-3-倍</link><generator>RSS for Node</generator><lastBuildDate>Mon, 18 May 2026 20:30:13 GMT</lastBuildDate><atom:link href="https://welinux.com//topic/22.rss" rel="self" type="application/rss+xml"/><pubDate>Wed, 22 Apr 2026 21:54:27 GMT</pubDate><ttl>60</ttl></channel></rss>