Huawei Ascend 910B vs Huawei Atlas 350

The Atlas 350 delivers 6% more FP16 throughput than the Ascend 910B (425 vs 400 TFLOPS dense). The Atlas 350 also carries 48 GB more memory (112 GB vs 64 GB).

Ascend 910B: Da Vinci v2, 2024 Atlas 350: Da Vinci v3, 2026

FP16 dense lead

+6%

Atlas 350: 425 vs 400 TFLOPS

Memory

64 vs 112 GB

48 GB more for the Atlas 350

Bandwidth

1.6 TB/s vs 1.4 TB/s

Ascend 910B moves data faster

TDP

400 W vs 600 W

Ascend 910B draws 200 W less

Download comparison sheet Printable version →

Huawei Ascend 910B

Full specs →

What is the Huawei Ascend 910B?

The Ascend 910B is an NPU from Huawei on the Da Vinci v2 architecture, an accelerator built for AI inference. It is an OAM module, the open accelerator format used in multi-GPU servers. Ascend 910B full specifications →

Can I rent the Ascend 910B in the cloud?

Flopper is not tracking any live rental listings for the Ascend 910B yet. The provider directory lists the clouds that rent comparable accelerators on demand. GPU cloud providers →

How does the Ascend 910B compare with other accelerators?

This page sets the Ascend 910B against the Atlas 350. The Ascend 910B is the older generation of the two. Both are Huawei parts. The compare tool builds the same comparison for any two parts in the catalogue. All GPU comparisons →

Where do the Ascend 910B figures come from?

Huawei published the Ascend 910B figures in marketing material rather than a datasheet, and Flopper has no independent confirmation of them. How Flopper verifies specifications →

Huawei Atlas 350

Leads on FP16 Full specs →

What is the Huawei Atlas 350?

The Atlas 350 is an NPU from Huawei on the Da Vinci v3 architecture, an accelerator built for AI inference. Atlas 350 full specifications →

Can I rent the Atlas 350 in the cloud?

Flopper is not tracking any live rental listings for the Atlas 350 yet. The provider directory lists the clouds that rent comparable accelerators on demand. GPU cloud providers →

How does the Atlas 350 compare with other accelerators?

This page sets the Atlas 350 against the Ascend 910B. The Atlas 350 is the newer generation of the two. Both are Huawei parts. The compare tool builds the same comparison for any two parts in the catalogue. All GPU comparisons →

Where do the Atlas 350 figures come from?

Every Atlas 350 figure on this page traces to Huawei's own datasheet or specification table. How Flopper verifies specifications →

Specifications

Ascend 910B

Atlas 350

Architecture

Da Vinci v2Da Vinci v3

Launch Year

20242026

Form Factor

OAMCustom

Memory

64 GB112 GB

Memory Bandwidth

1.6 TB/s1.4 TB/s

TDP

400 W600 W

Process Node

SMIC 7nm—

Spec

Ascend 910B

Atlas 350

Architecture Da Vinci v2Da Vinci v3

Launch Year 20242026

Form Factor OAMCustom

Memory 64 GB112 GB

Memory Bandwidth 1.6 TB/s1.4 TB/s

TDP 400 W600 W

Process Node SMIC 7nm—

Performance (TFLOPS)

Ascend 910B

Atlas 350

400 TFLOPS 425 TFLOPS

804 TFLOPS

FP6

804 TFLOPS

1,561 TFLOPS

800 TOPS 804 TOPS

Precision

Ascend 910B

Atlas 350

400 TFLOPS 425 TFLOPS

804 TFLOPS

FP6 804 TFLOPS

1,561 TFLOPS

800 TOPS 804 TOPS

Figures are dense. Where a vendor publishes a figure it sits under the dense one and never decides a winner. Hover any precision for its definition, or read the GPU glossary.

FLOPS by Precision

What actually differs

The Atlas 350 is the newer part: Da Vinci v3 (2026) against the Ascend 910B's Da Vinci v2 (2024).

The Atlas 350 carries 48 GB more memory (112 GB against 64 GB), so larger models fit on one GPU before you shard. Bandwidth is 1.6 TB/s against 1.4 TB/s, which sets the ceiling on decode speed.

Weight the FP8 and FP16 rows for LLM work and the FP64 row for HPC.

New to the terms? The glossary explains BF16, FP16, FP8, FP4, INT8, Sparsity and TFLOPS.

Frequently asked questions

What is better, the Ascend 910B or the Atlas 350?

The Atlas 350, on throughput. At FP16 precision the Atlas 350 reaches 425 TFLOPS dense against 400 TFLOPS for the Ascend 910B. The performance table on this page lists every published precision for both GPUs.

Which should you choose, the Ascend 910B or the Atlas 350?

Choose the Atlas 350 for training and high-throughput inference: it has 6% more FP16 throughput than the Ascend 910B. It also carries 48 GB more memory, so larger models fit on one GPU before you shard. The Atlas 350 draws 200 W more power, so budget for cooling and power delivery before you commit.

Which has more memory, the Ascend 910B or the Atlas 350?

The Atlas 350 carries 112 GB of memory versus 64 GB for the Ascend 910B. Memory bandwidth is 1.6 TB/s for the Ascend 910B and 1.4 TB/s for the Atlas 350.

How much power do the Ascend 910B and the Atlas 350 draw?

The Ascend 910B is rated at 400 W TDP and the Atlas 350 at 600 W.

Get Comparison Updates

New GPUs added weekly. Be the first to see how they compare.

Open in Full Compare Tool → Ascend 910B full specs Atlas 350 full specs Estimate cluster cost in the calculator →

Huawei Ascend 910B vs Huawei Atlas 350

Huawei Ascend 910B

What is the Huawei Ascend 910B?

Can I rent the Ascend 910B in the cloud?

How does the Ascend 910B compare with other accelerators?

Where do the Ascend 910B figures come from?

Huawei Atlas 350

What is the Huawei Atlas 350?

Can I rent the Atlas 350 in the cloud?

How does the Atlas 350 compare with other accelerators?

Where do the Atlas 350 figures come from?

Specifications

Performance (TFLOPS)

FLOPS by Precision

What actually differs

Frequently asked questions

What is better, the Ascend 910B or the Atlas 350?

Which should you choose, the Ascend 910B or the Atlas 350?

Which has more memory, the Ascend 910B or the Atlas 350?

How much power do the Ascend 910B and the Atlas 350 draw?

Get Comparison Updates

Stay Updated on GPU Specs