Cloud TPU generations

Tensor Processing Units (TPUs) are accelerators for machine learning. This set documents four generations, which share most concepts and tasks but differ in their values.

Each generation has an introduction, a specifications topic, a configurations topic and a VM types topic. Values such as chips per pod are stored once as keys, so a corrected value changes every page that shows it.

• TPU v5e

• TPU v5p

• TPU v6e (Trillium)

• TPU7x (Ironwood)

this topic set was checked against the published pages on 24 October 2026.

• Compare TPU v5p, TPU v6e (Trillium) and TPU7x (Ironwood)
  Side-by-side specifications for the three most recent generations, as published on the TPU7x (Ironwood) page.