Infrastructure atlas
Google/Custom ASIC

Cloud TPU v5p

Google’s large-scale training TPU with 3D-torus slices and enough fabric to schedule jobs across thousands of chips.

TPUTPUtrainingJAX3D torus
Cloud TPU v5p hardware

01 / Memory

95 GiB HBM

Per chip

02 / Bandwidth

2,765 GB/s

HBM bandwidth

03 / Pod

8,960 chips

3D torus

Precision & peak compute

PrecisionPublished peakBasis
BF16459 TFLOPSPer chip

Peak FLOPS and TOPS are theoretical hardware figures. Sparse and dense modes, chip and system totals, and form factors are not interchangeable.

Google Cloud TPU v5p documentation

Technical specification

Chip & memory
Peak BF16
459 TFLOPS per chip
TensorCores
2 per chip · 4 MXUs each
HBM
95 GiB · 2,765 GB/s
SparseCore
4 per chip
Pod & interconnect
Pod size
8,960 chips
ICI
1,200 GB/s bidirectional per chip
Topology
3D torus; twisted slices supported
Software
JAX, PyTorch/XLA, TensorFlow