DEPLOYDatabase

Chip comparison

Google TPU v4 vs Google TPU v5p

Side-by-side specs

Straight from each vendor's datasheet. Winner in bold: higher throughput, memory, or transistor count; lower power draw or price. Sparse (2:4) throughput roughly doubles the FP8/INT8 numbers shown.

FieldGoogle TPU v4Google TPU v5p
Process nodeTSMC N7
TDP192 W300 W
Memory32 GB HBM295 GB HBM2e
Memory bandwidth1,200 GB/s2,765 GB/s
BF16 (dense)275 TFLOPS459 TFLOPS
INT8 (dense)275 TOPS918 TOPS
Form factorOAM (per-chip in v4 pod)OAM (per-chip in v5p pod)
Announced2021-05-182023-12-06
Released2022-05-012024-01-01

Robots running Google TPU v4

No robots publicly running it yet.

Robots running Google TPU v5p

No robots publicly running it yet.

Data centers with Google TPU v5p

Common questions

Google TPU v4 vs Google TPU v5p: which is faster for training?

Google TPU v5p has 1.67x the dense FP16/BF16 throughput of the other (Google TPU v4: 275 TFLOPS; Google TPU v5p: 459 TFLOPS). Real-world training speed also depends on memory bandwidth, interconnect, and software stack: see the spec diff for those inputs.

Google TPU v4 vs Google TPU v5p: which has more memory?

Google TPU v5p carries more HBM (Google TPU v4: 32 GB HBM2; Google TPU v5p: 95 GB HBM2e). Memory capacity sets the largest model that fits on one chip without partitioning; memory bandwidth sets inference throughput.

Google TPU v4 vs Google TPU v5p: which is more power-efficient?

Dense FP16/BF16 TFLOPS per watt: Google TPU v4 1.43 (275 ÷ 192 W); Google TPU v5p 1.53 (459 ÷ 300 W). Google TPU v5p wins at the die level; system-level efficiency also depends on cooling and interconnect.

Google TPU v4 vs Google TPU v5p: which is more widely used?

Google TPU v4: 0 robots and 2 data centers publicly running it. Google TPU v5p: 0 robots and 1 data center. Empty here means no operator has said publicly, not zero adoption in the market.

See also: every chip comparison · Google TPU v4 full page · Google TPU v5p full page.