Huawei logo

Huawei Ascend 910A

Type: NPUArchitecture: Da VinciForm factor: OAMReleased: 2019Process: TSMC 7nmSpec confidence: Official
FP32
16
TFLOPS
Memory
32 GB
HBM2
Bandwidth
1.2 TB/s
memory
TDP
310 W
51.6 FP32 dense TFLOPS/kW of TDP

Overview

Huawei Ascend 910 (retroactively referred to as 910A after the 910B/910C launches) is Huawei's first-generation Da Vinci datacenter AI accelerator, launched in 2019 on TSMC's 7nm process. The chip packs 32 Da Vinci AI cores and delivers 256 TFLOPS FP16 / 512 TOPS INT8, paired with 32 GB HBM2 at 1.2 TB/s. It ships in the Atlas 800 server and Atlas 300T accelerator card, with 720 Gbps of HCCS inter-chip bandwidth plus 2x100 Gbps RoCE for scale-out. FP32 throughput is derived as approximately 1/16 of FP16. Third-party comparison tables sometimes list 4,096 'TPP' for this chip - this is the US BIS export-control metric (FP16 TFLOPS x 16 bits), not a TFLOPS figure.

Performance

Peak theoretical throughput by precision type

PrecisionPeak
FP64
No verified data available
FP32
32-bit floating point
16TFLOPS
TF32
No verified data available
BF16
No verified data available
FP16
16-bit floating point
256TFLOPS
FP8
No verified data available
FP6
No verified data available
FP4
No verified data available
INT8
8-bit integer
512TOPS

The Ascend 910A in the GPU landscape

Peak FP16 TFLOPS (dense) against TDP, single-GPU parts tracked by Flopper

06001,2001,8002,4003,0000 W250 W500 W750 W1000 W1250 W1500 WInstinct MI355XAscend 910A

Higher and further left is better: more half-precision throughput for less power.

Specifications

Architecture

Da Vinci

Form Factor

OAM

Launch Year

2019

Process Node

TSMC 7nm

Memory

32 GB HBM2

Bandwidth

1,228 GB/s

TDP

310 W

Max power (Flopper estimate)

~357 W est. Flopper estimate: 310 W TDP x 1.15. The vendor publishes no maximum board power for this part.

Interconnect

90 GB/s HCCS

direction and scope not stated by vendor

Spec Confidence

Official

Full Specifications

Memory
Memory 32 GB
Memory Type HBM2
Bandwidth 1.2 TB/s
Interface Width No verified data available
Interconnect & I/O
GPU-to-GPU HCCS
Interconnect Bandwidth 90 GB/s direction and scope not stated by vendor
Power & Thermal
TDP 310 W
Max power (Flopper estimate) ~357 W est.
General
Form Factor OAM
Architecture Da Vinci
Process Node TSMC 7nm
Launch Year 2019

Datasheet & Resources

Data Provenance

Every figure traced to a source

Primary Source

Publisher
SemiAnalysis
Published
2025-09-19

Data Quality

Spec confidence
Official
Clock basis
Boost
Core precisions with figures
3 of 9
Normalization
All values in TFLOPS

Compare cloud providers offering on-demand GPU instances for AI training, inference, and HPC workloads.

Browse GPU Cloud Providers

Similar GPUs

Huawei Atlas 300I Duo 48GB

PCIe · 2022
FP32: No verified data available
Compare vs Atlas 300I Duo

Huawei Atlas 300I Duo 96GB

PCIe · 2022
FP32: No verified data available
Compare vs Atlas 300I Duo

Huawei Atlas 300V Pro

PCIe · 2022
FP32: No verified data available
Compare vs Atlas 300V Pro

Huawei Atlas 300I Pro

PCIe · 2021
FP32: No verified data available
Compare vs Atlas 300I Pro

Huawei Atlas 300V

PCIe · 2022
FP32: No verified data available
Compare vs Atlas 300V

Frequently Asked Questions

How many TFLOPS does the Huawei Ascend 910A have?

The Huawei Ascend 910A delivers 16 TFLOPS FP32 and 256 TFLOPS FP16 at peak. Flopper does not currently have a verified FP8 throughput figure for it.

What is the power consumption of the Huawei Ascend 910A?

The Huawei Ascend 910A has a TDP (Thermal Design Power) rating of 310 watts.

How much memory does the Huawei Ascend 910A have?

The Huawei Ascend 910A is equipped with 32 GB of memory with 1,228 GB/s of memory bandwidth.

What architecture is the Huawei Ascend 910A based on?

The Huawei Ascend 910A is based on the Da Vinci architecture, launched in 2019.

Stay Updated on GPU Releases

Get notified when new GPUs are added or specifications are updated.

Loading verification...

No spam, unsubscribe anytime.

Back to GPUs

© 2026 Flopper.io - Compare the hardware powering AI