DEPLOYDatabase

Chip comparison

Google TPU v4 vs Huawei Ascend 910B

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 v4Huawei Ascend 910B
Process nodeTSMC N7SMIC N+2 (~7nm)
TDP192 W400 W
Memory32 GB HBM264 GB HBM2e
Memory bandwidth1,200 GB/s1,600 GB/s
FP16 (dense)400 TFLOPS
BF16 (dense)275 TFLOPS400 TFLOPS
INT8 (dense)275 TOPS800 TOPS
Form factorOAM (per-chip in v4 pod)OAM (Atlas 300T A2)
Announced2021-05-182023-08-01
Released2022-05-012023-08-01

Robots running Google TPU v4

No robots publicly running it yet.

Robots running Huawei Ascend 910B

Data centers with Huawei Ascend 910B

Common questions

Google TPU v4 vs Huawei Ascend 910B: which is faster for training?

Huawei Ascend 910B has 1.45x the dense FP16/BF16 throughput of the other (Google TPU v4: 275 TFLOPS; Huawei Ascend 910B: 400 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 Huawei Ascend 910B: which has more memory?

Huawei Ascend 910B carries more HBM (Google TPU v4: 32 GB HBM2; Huawei Ascend 910B: 64 GB HBM2e). Memory capacity sets the largest model that fits on one chip without partitioning; memory bandwidth sets inference throughput.

Google TPU v4 vs Huawei Ascend 910B: which is more power-efficient?

Dense FP16/BF16 TFLOPS per watt: Google TPU v4 1.43 (275 ÷ 192 W); Huawei Ascend 910B 1.00 (400 ÷ 400 W). Google TPU v4 wins at the die level; system-level efficiency also depends on cooling and interconnect.

Google TPU v4 vs Huawei Ascend 910B: which is more widely used?

Google TPU v4: 0 robots and 2 data centers publicly running it. Huawei Ascend 910B: 2 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 · Huawei Ascend 910B full page.