Flopper.io Flopper.io
Huawei logo

Huawei Ascend 910A vs Huawei Atlas 350

The Atlas 350 delivers 66% more FP16 throughput than the Ascend 910A (425 vs 256 TFLOPS dense). The Atlas 350 also carries 80 GB more memory (112 GB vs 32 GB).

Ascend 910A: Da Vinci, 2019 Atlas 350: Da Vinci v3, 2026
FP16 dense lead
+66%
Atlas 350: 425 vs 256 TFLOPS
Memory
32 vs 112 GB
80 GB more for the Atlas 350
Bandwidth
1.2 TB/s vs 1.4 TB/s
Atlas 350 moves data faster
TDP
310 W vs 600 W
Ascend 910A draws 290 W less
Download comparison sheet

Share this page

Huawei Ascend 910A 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 910A

Full specs →

What is the Huawei Ascend 910A?

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

Can I rent the Ascend 910A in the cloud?

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

How does the Ascend 910A compare with other accelerators?

This page sets the Ascend 910A against the Atlas 350. The Ascend 910A 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 910A figures come from?

Every Ascend 910A figure on this page traces to Huawei's own datasheet or specification table. How Flopper verifies specifications →

Huawei logo

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 910A. 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

Huawei logo
Ascend 910A
Huawei logo
Atlas 350
Architecture
Da VinciDa Vinci v3
Launch Year
20192026
Form Factor
OAMCustom
Memory
32 GB112 GB
Memory Bandwidth
1.2 TB/s1.4 TB/s
TDP
310 W600 W
Process Node
TSMC 7nm—

Performance (TFLOPS)

Huawei logo
Ascend 910A
Huawei logo
Atlas 350
No verified data available No verified data available
16 TFLOPS No verified data available
No verified data available No verified data available
No verified data available425 TFLOPS
256 TFLOPS 425 TFLOPS
No verified data available804 TFLOPS
FP6
No verified data available804 TFLOPS
No verified data available1,561 TFLOPS
512 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 910A's Da Vinci (2019).

The Atlas 350 carries 80 GB more memory (112 GB against 32 GB), so larger models fit on one GPU before you shard. Bandwidth is 1.2 TB/s against 1.4 TB/s, which sets the ceiling on decode speed.

Power is 310 W against 600 W, and at FP32 the Ascend 910A does more work per watt. Weight the FP8 and FP16 rows for LLM work and the FP64 row for HPC.

New to the terms? The glossary explains FP32, BF16, FP16, FP8, FP4, INT8, Sparsity and TFLOPS.

Frequently asked questions

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

The Atlas 350, on throughput. At FP16 precision the Atlas 350 reaches 425 TFLOPS dense against 256 TFLOPS for the Ascend 910A. The performance table on this page lists every published precision for both GPUs.

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

Choose the Atlas 350 for training and high-throughput inference: it has 66% more FP16 throughput than the Ascend 910A. It also carries 80 GB more memory, so larger models fit on one GPU before you shard. The Atlas 350 draws 290 W more power, so budget for cooling and power delivery before you commit.

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

The Atlas 350 carries 112 GB of memory versus 32 GB for the Ascend 910A. Memory bandwidth is 1.2 TB/s for the Ascend 910A and 1.4 TB/s for the Atlas 350.

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

The Ascend 910A is rated at 310 W TDP and the Atlas 350 at 600 W. On FP32 throughput per watt, the Ascend 910A is the more efficient part.

Get Comparison Updates

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

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

© 2026 Flopper.io - Compare the GPUs Powering AI