inference/research
verifiedhigh · 0.95updated 2026-10-04

Cloud TPU performance guide

spec-sheet

Fields

id
sources/alt-tpu-perf-guide
Publisher
Google Cloud
Kind
spec-sheet
Published
2026-10-02
Accessed
2026-10-04
notes
XLA compilation model, padding to 128x8 chunks, batch size rules (multiple of 64/1024), fusion, broadcasting, MXU size 256x256 for v6e+, memory management (TPU_PREMAPPED_BUFFER_SIZE, pinned_host vs unpinned_host), MTU 8896, multi-NIC, Ironwood dual-chiplet recommendations.

Notes

XLA compilation model, padding to 128x8 chunks, batch size rules (multiple of 64/1024), fusion, broadcasting, MXU size 256x256 for v6e+, memory management (TPU_PREMAPPED_BUFFER_SIZE, pinned_host vs unpinned_host), MTU 8896, multi-NIC, Ironwood dual-chiplet recommendations.

Cross-references

This record references no other records.