DEPLOYDatabase

Chip comparison

Huawei Ascend 910B vs NVIDIA B100 (Blackwell)

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.

FieldHuawei Ascend 910BNVIDIA B100 (Blackwell)
Process nodeSMIC N+2 (~7nm)TSMC 4NP
TDP400 W700 W
Memory64 GB HBM2e192 GB HBM3e
Memory bandwidth1,600 GB/s8,000 GB/s
FP16 (dense)400 TFLOPS1,750 TFLOPS
BF16 (dense)400 TFLOPS1,750 TFLOPS
FP8 (dense)3,500 TFLOPS
INT8 (dense)800 TOPS
Form factorOAM (Atlas 300T A2)SXM6 (700 W envelope)
Announced2023-08-012024-03-18
Released2023-08-012024-11-01

Robots running Huawei Ascend 910B

Robots running NVIDIA B100 (Blackwell)

No robots publicly running it yet.

Data centers with Huawei Ascend 910B

Data centers with NVIDIA B100 (Blackwell)

No data centers publicly running it yet.

Common questions

Huawei Ascend 910B vs NVIDIA B100 (Blackwell): which is faster for training?

NVIDIA B100 (Blackwell) has 8.75x the dense FP16/BF16 throughput of the other (Huawei Ascend 910B: 400 TFLOPS; NVIDIA B100 (Blackwell): 3,500 TFLOPS). Real-world training speed also depends on memory bandwidth, interconnect, and software stack: see the spec diff for those inputs.

Huawei Ascend 910B vs NVIDIA B100 (Blackwell): which has more memory?

NVIDIA B100 (Blackwell) carries more HBM (Huawei Ascend 910B: 64 GB HBM2e; NVIDIA B100 (Blackwell): 192 GB HBM3e). Memory capacity sets the largest model that fits on one chip without partitioning; memory bandwidth sets inference throughput.

Huawei Ascend 910B vs NVIDIA B100 (Blackwell): which is more power-efficient?

Dense FP16/BF16 TFLOPS per watt: Huawei Ascend 910B 1.00 (400 ÷ 400 W); NVIDIA B100 (Blackwell) 5.00 (3,500 ÷ 700 W). NVIDIA B100 (Blackwell) wins at the die level; system-level efficiency also depends on cooling and interconnect.

Huawei Ascend 910B vs NVIDIA B100 (Blackwell): which is more widely used?

Huawei Ascend 910B: 2 robots and 1 data center publicly running it. NVIDIA B100 (Blackwell): 0 robots and 0 data centers. Empty here means no operator has said publicly, not zero adoption in the market.

See also: every chip comparison · Huawei Ascend 910B full page · NVIDIA B100 (Blackwell) full page.