AI Infrastructure DatabaseAI背后的芯片、电力与企业
已保存
菜单
首页 如何构建 关注清单 The Stack
硬件 芯片芯片对比 服务器与系统网络互联 内存晶圆厂与封装 冷却
设施 数据中心电力 项目云计算与 GPU 算力
智能 AI 模型软件 AI 产品基准测试 研究论文数据集 概念
生态系统 公司股票 人物时间线 趋势AI 的未来
学习 新闻 询问数据 AI 智能体
关于 关于我们 研究方法 数据与 API 免责声明
选项
数据与 API

免费只读 JSON 接口,访问本站缓存数据。

深色模式

引导视图
对芯片、数据中心和模型相关术语感到陌生?我们在您浏览时以简明语言解释关键概念,数据不变,帮助随时在线。

专家视角
您已了解整体架构,这里只呈现数据:简洁、快速、紧凑,不附加额外说明。这是默认视图。

界面语言
‹ 返回
首页 › 芯片 › Google TPU 8t / TPU 8i

Google TPU 8t / TPU 8i

芯片 AI 加速器 已宣布

Google's eighth TPU generation, split for the first time into a training chip and an inference chip.

您正在查看一条chip记录:包括该处理器的定义、核心规格、设计方、制造商、零部件供应商,以及其应用场景。

Announced at Google Cloud Next on April 22, 2026, the eighth TPU generation comes as two chips. TPU 8t targets training: 12.6 PFLOPS of FP4 and 216 GB of HBM per chip, in 9,600-chip superpods with about 2 PB of shared memory. TPU 8i targets serving and reinforcement learning: 10.1 PFLOPS of FP4, 288 GB of HBM and 384 MB of on-chip SRAM, linked by a new Boardfly topology. Google claims up to 2x performance per watt over Ironwood. In November 2026 it added TPU 8tx (confidential computing) and TPU 8i Max (two 8i dies per package); a general availability date is not confirmed.

规格参数与关键信息

每项数据均标注来源依据:厂商规格、厂商声明、媒体报道、估算或官方数据。"未公开披露"即字面含义,我们从不编造数值。

Overview

Designer Google
Announced April 22, 2026 (Google Cloud Next)Official Google Cloud deep dive, 2026-04-22
Status Announced: GA 'later this year' per Apr 2026 launch; no GA date confirmed as of Nov 2026 blog.google, Apr 2026; Cloud TPU release notes list no TPU 8 entry
Variants TPU 8tx (8t silicon, AMD EPYC hosts, confidential compute); TPU 8i Max (two 8i chips per package)Official blog.google, 2026-11-09

算力

TPU 8t FP4 (peak) 12.6 PFLOPSVendor spec Per chip; Google Cloud deep dive, Apr 2026
TPU 8i FP4 (peak) 10.1 PFLOPSVendor spec Per chip; Google Cloud deep dive, Apr 2026

内存

TPU 8t HBM 216 GB, 6,528 GB/sVendor spec Per chip; Google Cloud deep dive
TPU 8i HBM 288 GB, 8,601 GB/sVendor spec Per chip; Google Cloud deep dive
On-chip SRAM 128 MB (8t); 384 MB (8i)Vendor spec Google Cloud deep dive

Physical

TPU 8t superpod 9,600 chips; ~2 PB shared HBM; 121 ExaFLOPSVendor claim blog.google, Apr 2026
TPU 8i pod Up to 1,152 chips in Boardfly topology; ICI 19.2 Tb/sVendor claim Google Cloud Next '26 post and blog.google, Apr 2026
Scale-out Virgo fabric: 134,000+ TPU 8t chips per data center; 1M+ across sitesVendor claim Google Cloud, Apr 2026
Host / cooling Axion Arm CPU hosts; fourth-generation liquid coolingVendor spec blog.google, Apr 2026

电力

Efficiency Up to 2x performance per watt vs IronwoodVendor claim Google, Apr 2026

Manufacturing

Process Not publicly disclosed Not disclosed by Google; press reports of TSMC 2 nm unconfirmed

与技术栈其他层级的关联

本条目关联一览:点击分组可直接跳转。

设计方

内部组成

使用

发布渠道

数据来源

Google Cloud: TPU 8t and TPU 8i technical deep dive (Apr 2026) ↗ · Google: Our eighth generation TPUs (Apr 2026) ↗ · Google Cloud: AI infrastructure at Next '26 (Apr 2026) ↗ · Google: Introducing TPU 8tx and TPU 8i Max (Nov 2026) ↗

记录审核时间 2026-11-16; 查看 研究方法.