返回
原文
Rohan Paul@rohanpaul_ai
36AI 编辑部评分，满分 100
2026-08-06 01:33· 23分钟前
AI 导读

Jeff Dean 在访谈中回顾 TPU 的起源：2010 年代初他估算，若 1 亿人每天用语音识别 3 分钟且全跑在 CPU 上，谷歌需为此增加约一倍服务器，成本与功耗不可接受。他因此力排众议，放弃 FPGA，推动定制 ASIC 芯片 TPUv1（PCIe 卡上的推理芯片），并说服 CFO 在用例完全明确前先部署约 5000 万美元的该芯片。

Here, Jeff Dean (Google's chief scientist) explains here how the idea behind TPUs was born at Google.

In the early 2010s, he noticed that new speech recognition and vision were getting much better, but they were extremely heavy to run on regular CPUs.

His back-of-the-envelope calculation was: if speech recognition got good enough that, say, 100M people talked to their phone for 3 minutes a day, and all of that ran on CPU servers, Google might need roughly 2x the total number of machines it already had, just for that one feature.

That was obviously insane in cost, power, and time to deploy, so he concluded they needed a completely different kind of hardware, purpose built for neural nets.

Also given neural nets mostly use a small set of simple numeric operations and are very tolerant of low precision and noise, so specialized chips could be dramatically more efficient than CPUs or GPUs.

Instead of using flexible but inefficient FPGAs, he pushed for a custom ASIC: TPUv1, an inference chip on a PCIe card, then convinced the CFO to deploy about $50M worth of them in data centers before every use case was even fully mapped out.

---

From 'Radical Ventures' YT channel (link in comment)

Rohan PaulGoogle's chief scientist is leaving after 27 years to build AI that can run its own research cycle with little human help. Jeff Dean was employee 30, and he wro...
Google教程/实践部署/工程

来源：Rohan Paul · x.com

导出 Markdown
返回
36在 X 查看原推
导出 Markdown
Rohan Paul · @rohanpaul_ai · X·2026-08-06 01:33·23分钟前
AI 导读

Jeff Dean 在访谈中回顾 TPU 的起源：2010 年代初他估算，若 1 亿人每天用语音识别 3 分钟且全跑在 CPU 上，谷歌需为此增加约一倍服务器，成本与功耗不可接受。他因此力排众议，放弃 FPGA，推动定制 ASIC 芯片 TPUv1（PCIe 卡上的推理芯片），并说服 CFO 在用例完全明确前先部署约 5000 万美元的该芯片。

Here, Jeff Dean (Google's chief scientist) explains here how the idea behind TPUs was born at Google.

In the early 2010s, he noticed that new speech recognition and vision were getting much better, but they were extremely heavy to run on regular CPUs.

His back-of-the-envelope calculation was: if speech recognition got good enough that, say, 100M people talked to their phone for 3 minutes a day, and all of that ran on CPU servers, Google might need roughly 2x the total number of machines it already had, just for that one feature.

That was obviously insane in cost, power, and time to deploy, so he concluded they needed a completely different kind of hardware, purpose built for neural nets.

Also given neural nets mostly use a small set of simple numeric operations and are very tolerant of low precision and noise, so specialized chips could be dramatically more efficient than CPUs or GPUs.

Instead of using flexible but inefficient FPGAs, he pushed for a custom ASIC: TPUv1, an inference chip on a PCIe card, then convinced the CFO to deploy about $50M worth of them in data centers before every use case was even fully mapped out.

---

From 'Radical Ventures' YT channel (link in comment)

Rohan PaulGoogle's chief scientist is leaving after 27 years to build AI that can run its own research cycle with little human help. Jeff Dean was employee 30, and he wro...

来源：Rohan Paul· x.com

Google教程/实践部署/工程