Huawei Ascend 950PR vs Huawei Atlas 350

The Ascend 950PR delivers 24% more FP8 throughput than the Atlas 350 (1,000 vs 804 TFLOPS dense). The Ascend 950PR also carries 16 GB more memory (128 GB vs 112 GB).

Ascend 950PR: Da Vinci v3, 2026 Atlas 350: Da Vinci v3, 2026

FP8 dense lead

+24%

Ascend 950PR: 1,000 vs 804 TFLOPS

Memory

128 vs 112 GB

16 GB more for the Ascend 950PR

Bandwidth

1.6 TB/s vs 1.4 TB/s

Ascend 950PR moves data faster

TDP

900 W vs 600 W

Atlas 350 draws 300 W less

Download comparison sheet Printable version →

Huawei Ascend 950PR

Leads on FP8 Full specs →

What is the Huawei Ascend 950PR?

The Ascend 950PR is an NPU from Huawei on the Da Vinci v3 architecture, an accelerator built for AI inference. It is an OAM module, the open accelerator format used in multi-GPU servers. Ascend 950PR full specifications →

Can I rent the Ascend 950PR in the cloud?

Flopper is not tracking any live rental listings for the Ascend 950PR yet. The provider directory lists the clouds that rent comparable accelerators on demand. GPU cloud providers →

How does the Ascend 950PR compare with other accelerators?

This page sets the Ascend 950PR against the Atlas 350. Both parts launched in the same year. 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 950PR figures come from?

Huawei published the Ascend 950PR figures in marketing material rather than a datasheet, and Flopper has no independent confirmation of them. How Flopper verifies specifications →

Huawei Atlas 350

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 950PR. Both parts launched in the same year. 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 950PR

Atlas 350

Architecture

Da Vinci v3Da Vinci v3

Launch Year

20262026

Form Factor

OAMCustom

Memory

128 GB112 GB

Memory Bandwidth

1.6 TB/s1.4 TB/s

TDP

900 W600 W

Process Node

——

Spec

Ascend 950PR

Atlas 350

Architecture Da Vinci v3Da Vinci v3

Launch Year 20262026

Form Factor OAMCustom

Memory 128 GB112 GB

Memory Bandwidth 1.6 TB/s1.4 TB/s

TDP 900 W600 W

Process Node ——

Performance (TFLOPS)

Ascend 950PR

Atlas 350

500 TFLOPS 425 TFLOPS

1,000 TFLOPS 804 TFLOPS

FP6

804 TFLOPS

2,000 TFLOPS 1,561 TFLOPS

804 TOPS

Precision

Ascend 950PR

Atlas 350

500 TFLOPS 425 TFLOPS

1,000 TFLOPS 804 TFLOPS

FP6 804 TFLOPS

2,000 TFLOPS 1,561 TFLOPS

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

Both launched in 2026, the Ascend 950PR on Da Vinci v3 and the Atlas 350 on Da Vinci v3.

The Ascend 950PR carries 16 GB more memory (128 GB against 112 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

Is the Ascend 950PR faster than the Atlas 350?

At FP8 precision the Ascend 950PR reaches 1,000 TFLOPS dense against 804 TFLOPS for the Atlas 350. The performance table on this page lists every published precision for both GPUs.

Which should you choose, the Ascend 950PR or the Atlas 350?

Choose the Ascend 950PR for training and high-throughput inference: it has 24% more FP8 throughput than the Atlas 350. It also carries 16 GB more memory, so larger models fit on one GPU before you shard. The Ascend 950PR draws 300 W more power, so budget for cooling and power delivery before you commit.

Which has more memory, the Ascend 950PR or the Atlas 350?

The Ascend 950PR carries 128 GB of memory versus 112 GB for the Atlas 350. Memory bandwidth is 1.6 TB/s for the Ascend 950PR and 1.4 TB/s for the Atlas 350.

How much power do the Ascend 950PR and the Atlas 350 draw?

The Ascend 950PR is rated at 900 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 950PR full specs Atlas 350 full specs Estimate cluster cost in the calculator →

Huawei Ascend 950PR vs Huawei Atlas 350

Huawei Ascend 950PR

What is the Huawei Ascend 950PR?

Can I rent the Ascend 950PR in the cloud?

How does the Ascend 950PR compare with other accelerators?

Where do the Ascend 950PR 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

Is the Ascend 950PR faster than the Atlas 350?

Which should you choose, the Ascend 950PR or the Atlas 350?

Which has more memory, the Ascend 950PR or the Atlas 350?

How much power do the Ascend 950PR and the Atlas 350 draw?

Get Comparison Updates

Stay Updated on GPU Specs