DEPLOYDatabase

Chip comparison

AWS Inferentia 2 vs Google TPU v4

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.

FieldAWS Inferentia 2Google TPU v4
Process nodeTSMC N7TSMC N7
TDP250 W192 W
Memory32 GB HBM332 GB HBM2
Memory bandwidth820 GB/s1,200 GB/s
BF16 (dense)190 TFLOPS275 TFLOPS
FP8 (dense)380 TFLOPS
INT8 (dense)275 TOPS
Form factorInf2 EC2 instanceOAM (per-chip in v4 pod)
Announced2022-12-012021-05-18
Released2023-04-012022-05-01

Robots running AWS Inferentia 2

No robots publicly running it yet.

Robots running Google TPU v4

No robots publicly running it yet.

Data centers with AWS Inferentia 2

No data centers publicly running it yet.

Common questions

AWS Inferentia 2 vs Google TPU v4: which is faster for training?

AWS Inferentia 2 has 1.38x the dense FP16/BF16 throughput of the other (AWS Inferentia 2: 380 TFLOPS; Google TPU v4: 275 TFLOPS). Real-world training speed also depends on memory bandwidth, interconnect, and software stack: see the spec diff for those inputs.

AWS Inferentia 2 vs Google TPU v4: which has more memory?

Google TPU v4 carries more HBM (AWS Inferentia 2: 32 GB HBM3; Google TPU v4: 32 GB HBM2). Memory capacity sets the largest model that fits on one chip without partitioning; memory bandwidth sets inference throughput.

AWS Inferentia 2 vs Google TPU v4: which is more power-efficient?

Dense FP16/BF16 TFLOPS per watt: AWS Inferentia 2 1.52 (380 ÷ 250 W); Google TPU v4 1.43 (275 ÷ 192 W). AWS Inferentia 2 wins at the die level; system-level efficiency also depends on cooling and interconnect.

AWS Inferentia 2 vs Google TPU v4: which is more widely used?

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

See also: every chip comparison · AWS Inferentia 2 full page · Google TPU v4 full page.