Subscribe

Google Tensor G5

The foundry switch to TSMC, and the first Tensor generation Google was willing to quantify properly.

Google stated the node for the first time in the family’s history, moving off Samsung Foundry to TSMC 3nm. A 60 per cent larger TPU let a newer, larger Gemini Nano run faster than the old one had, with a context window nearly tripled.

About Google All AI chips

Silicon from Google

13 chips on record, newest first.

  1. 2026

    Google Tensor G6

    3nm (TSMC) Tensor Processing Unit (TPU), co-designed with Google DeepMind

    50% more TPU compute than Tensor G5; on-device AI tasks up to 3.5x faster using up to 3.5x less energy

    Relative only
  2. 2026

    Google TPU 8i

    Not stated by Google Eighth-gen inference TPU, 384 MB on-chip SRAM

    Peak FP4 10.1 PFLOPs per chip

    Absolute
  3. 2026

    Google TPU 8t

    Not stated by Google Eighth-gen training TPU, up to 9,600 chips per superpod

    Peak FP4 12.6 PFLOPs per chip

    Absolute
  4. 2025

    Google Tensor G5

    3nm (TSMC) Tensor Processing Unit (TPU)

    Up to 60% more powerful TPU than Tensor G4; on-device Gemini Nano 2.6x faster and 2x more efficient; context window 12,000 to 32,000 tokens

    Relative only
  5. 2025

    Google TPU Ironwood (TPU7x)

    Not stated by Google Seventh-generation Tensor Processing Unit built for inference

    Each individual chip boasts peak compute of 4,614 TFLOPs.

    Absolute
  6. 2024

    Google Tensor G4

    Not stated by Google Tensor Processing Unit (TPU)

    No TPU figure published

    None published
  7. 2024

    Google TPU Trillium (v6e)

    Not stated by Google Sixth-generation Tensor Processing Unit

    918 TFLOPs BF16 per chip; 1,836 TOPs INT8

    Absolute
  8. 2023

    Google TPU v5p

    Not stated by Google Training-focused Tensor Processing Unit

    459 TFLOPs BF16 per chip

    Absolute
  9. 2023

    Google Tensor G3

    Not stated by Google Tensor Processing Unit (TPU)

    Runs more than twice as many on-device models as Pixel 6; on-device generative AI 150x more complex than Pixel 7’s largest model

    Relative only
  10. 2023

    Google TPU v5e

    Not stated by Google Cost-focused Tensor Processing Unit

    197 TFLOPs BF16 per chip; 393 TOPs INT8

    Absolute
  11. 2022

    Google Tensor G2

    Not stated by Google Tensor Processing Unit (TPU)

    No quantified claim published

    None published
  12. 2021

    Google Tensor

    Not stated by Google Tensor Processing Unit (TPU)

    No figure published; qualitative only

    None published
  13. 2021

    Google TPU v4

    Not stated by Google Google Tensor Processing Unit, fourth generation

    275 teraflops (bf16 or int8) per chip

    Absolute

Compare against every other vendor in the silicon tracker

Google in the news