In volume production and deployment since 2025, sold inside mainland China only, and the chip behind the CloudMatrix 384 supernode. Huawei is on the US Entity List and Ascend silicon is export-controlled wherever it is located, so it cannot be rented from any provider on this site. Huawei publishes no 910C datasheet, but its own CloudMatrix384 engineering paper (arXiv:2506.12708v1) documents the compute, core counts, memory and interconnect figures used here. The citation is pinned to version 1 of that paper on purpose. Huawei has since revised it twice: version 2 deletes the sentence giving the chip's throughput, and version 3 renames the part from Ascend 910C to Ascend 910 throughout. The figures recorded here were published by Huawei in version 1 and are no longer on the version the paper's main address serves, so anyone rechecking them needs the pinned link.
Huawei logo

Huawei Ascend 910C

Type: NPUArchitecture: Da Vinci v2 (dual-die)Form factor: OAMReleased: 2025Process: SMIC 7nmSpec confidence: Official
FP32
50
TFLOPS
Memory
128 GB
HBM2e
Bandwidth
3.2 TB/s
memory
TDP
600 W
83.3 FP32 dense TFLOPS/kW of TDP

Overview

The Huawei Ascend 910C is a dual-die package fusing two 910B-class dies on an organic substrate, mass shipping from Q1 2025. SemiAnalysis teardowns confirm 800 TFLOPS FP16 / 780 TFLOPS BF16 dense per chip, paired with 128 GB of HBM2e across 8 stacks delivering 3.2 TB/s of memory bandwidth (per Huawei Connect 2025 keynote). The chip supports FP32, HF32, FP16, BF16, and INT8 - but no native FP8 or FP4 (that arrives with the 950 generation). Scale-up bandwidth is 2,800 Gb/s over Huawei's UB (Unified Bus) fabric with 400 Gb/s of scale-out RDMA. The 910C is the foundation of the CloudMatrix 384 supercluster, where 384 chips deliver 300 PFLOPS BF16 dense across 49.2 TB of HBM at a system power of 559 kW.

Performance

Peak theoretical throughput by precision type

PrecisionPeak
FP64
No verified data available
FP32
32-bit floating point
50TFLOPS
TF32
No verified data available
BF16
Brain Float 16
800TFLOPS
FP16
16-bit floating point
800TFLOPS
FP8
No verified data available
FP6
No verified data available
FP4
No verified data available
INT8
8-bit integer
1,600TOPS

The Ascend 910C 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 910C

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

Specifications

Architecture

Da Vinci v2 (dual-die)

Form Factor

OAM

Launch Year

2025

Process Node

SMIC 7nm

Memory

128 GB HBM2e

Bandwidth

3,200 GB/s

TDP

600 W

Max power (Flopper estimate)

~690 W est. Flopper estimate: 600 W TDP x 1.15. The vendor publishes no maximum board power for this part.

Interconnect

350 GB/s Unified Bus (UB)

direction and scope not stated by vendor

Spec Confidence

Official

Full Specifications

Memory
Memory 128 GB
Memory Type HBM2e
Bandwidth 3.2 TB/s
Interface Width No verified data available
Interconnect & I/O
GPU-to-GPU Unified Bus (UB)
Interconnect Bandwidth 350 GB/s direction and scope not stated by vendor
Power & Thermal
TDP 600 W
Max power (Flopper estimate) ~690 W est.
General
Form Factor OAM
Architecture Da Vinci v2 (dual-die)
Process Node SMIC 7nm
Launch Year 2025

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
4 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

Systems Using This GPU

Pre-configured systems featuring the Huawei Ascend 910C

SystemGPU CountPeak PerformanceTotal Power
Huawei Atlas 800I A3
Atlas · rack · 2025
8x Ascend 910C 4.48 PFLOPS 14.6 kW View System
Huawei Atlas 800T A3
Atlas · rack · 2025
8x Ascend 910C 6.00 PFLOPS No verified data available View System
Huawei Atlas 900 A3 SuperPoD
Atlas · pod · 2025
384x Ascend 910C 614 PFLOPS No verified data available View System

Frequently Asked Questions

How many TFLOPS does the Huawei Ascend 910C have?

The Huawei Ascend 910C delivers 50 TFLOPS FP32 and 800 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 910C?

The Huawei Ascend 910C has a TDP (Thermal Design Power) rating of 600 watts.

How much memory does the Huawei Ascend 910C have?

The Huawei Ascend 910C is equipped with 128 GB of memory with 3,200 GB/s of memory bandwidth.

What architecture is the Huawei Ascend 910C based on?

The Huawei Ascend 910C is based on the Da Vinci v2 (dual-die) architecture, launched in 2025.

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