ModelRefs / TPU (Tensor Processing Unit) — AI Glossary
TPU (Tensor Processing Unit) — AI Glossary
Google's custom ASIC for matrix multiplication workloads, used to train and serve Gemini and other Google models. Also called Tensor Processing Unit.
Overview
Cloud TPU v5p (2024) offers 459 TFLOPS BF16 per chip with 95 GB HBM2e. TPUs excel at large-scale, homogeneous matrix ops and are the backbone of Google's ML infrastructure. Available via Cloud TPU and used in Google Colab. Architecture differs fundamentally from GPU: systolic array design with 2D data flow.
Reference details
| Topic | inference |
|---|---|
| Also known as | Tensor Processing Unit |
| Last reviewed | 2026-06-24 |
Related terms
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to TPU (Tensor Processing Unit) — AI Glossary.
Frequently asked questions
What is TPU (Tensor Processing Unit)?
Google's custom ASIC for matrix multiplication workloads, used to train and serve Gemini and other Google models.
Is TPU (Tensor Processing Unit) the same as Tensor Processing Unit?
Yes — Tensor Processing Unit are common aliases for TPU (Tensor Processing Unit).
What concepts are related to TPU (Tensor Processing Unit)?
Closely related concepts include h100, gpu memory, distributed inference.