Flopper.io Flopper.io
Huawei logo

Huawei Ascend 970 vs Huawei Atlas 350

The Ascend 970 delivers 5.0x the FP8 throughput of the Atlas 350 (4,000 vs 804 TFLOPS dense). The Ascend 970 also carries 176 GB more memory (288 GB vs 112 GB).

Ascend 970: Da Vinci v5, 2028 Atlas 350: Da Vinci v3, 2026
FP8 dense lead
5.0x
Ascend 970: 4,000 vs 804 TFLOPS
Memory
288 vs 112 GB
176 GB more for the Ascend 970
Bandwidth
14.4 TB/s vs 1.4 TB/s
Ascend 970 moves data faster
Download comparison sheet

Share this page

Huawei Ascend 970 vs Huawei Atlas 350

A post with this link shows the card on the Image tab as its preview.

Post on X Share on LinkedIn
 Printable version →
Huawei logo

Huawei Ascend 970

Leads on FP8 Full specs →

What is the Huawei Ascend 970?

The Ascend 970 is an NPU from Huawei on the Da Vinci v5 architecture, an accelerator built for AI inference. It is an OAM module, the open accelerator format used in multi-GPU servers. It has been announced but is not yet shipping. Ascend 970 full specifications →

Can I rent the Ascend 970 in the cloud?

Flopper is not tracking any live rental listings for the Ascend 970 yet. The provider directory lists the clouds that rent comparable accelerators on demand. GPU cloud providers →

How does the Ascend 970 compare with other accelerators?

This page sets the Ascend 970 against the Atlas 350. The Ascend 970 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 Ascend 970 figures come from?

Huawei published the Ascend 970 figures in marketing material rather than a datasheet, and Flopper has no independent confirmation of them. How Flopper verifies specifications →

Huawei logo

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 970. The Atlas 350 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 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

Huawei logo
Ascend 970
Huawei logo
Atlas 350
Architecture
Da Vinci v5Da Vinci v3
Launch Year
20282026
Form Factor
OAMCustom
Memory
288 GB112 GB
Memory Bandwidth
14.4 TB/s1.4 TB/s
TDP
—600 W
Process Node
——

Performance (TFLOPS)

Huawei logo
Ascend 970
Huawei logo
Atlas 350
No verified data available No verified data available
No verified data available No verified data available
No verified data available No verified data available
2,000 TFLOPS 425 TFLOPS
2,000 TFLOPS 425 TFLOPS
4,000 TFLOPS 804 TFLOPS
FP6
No verified data available804 TFLOPS
8,000 TFLOPS 1,561 TFLOPS
No verified data available804 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 Ascend 970 is the newer part: Da Vinci v5 (2028) against the Atlas 350's Da Vinci v3 (2026).

The Ascend 970 carries 176 GB more memory (288 GB against 112 GB), so larger models fit on one GPU before you shard. Bandwidth is 14.4 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 970 or the Atlas 350?

The Ascend 970, on throughput. At FP8 precision the Ascend 970 reaches 4,000 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 970 or the Atlas 350?

Choose the Ascend 970 for training and high-throughput inference: it has 5.0x the FP8 throughput of the Atlas 350. It also carries 176 GB more memory, so larger models fit on one GPU before you shard.

Which has more memory, the Ascend 970 or the Atlas 350?

The Ascend 970 carries 288 GB of memory versus 112 GB for the Atlas 350. Memory bandwidth is 14.4 TB/s for the Ascend 970 and 1.4 TB/s for the Atlas 350.

Get Comparison Updates

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

Open in Full Compare Tool → Ascend 970 full specs Atlas 350 full specs Estimate cluster cost in the calculator →

© 2026 Flopper.io - Compare the GPUs Powering AI