Sold in mainland China only. The Atlas 350 is distributed through Huawei's Chinese channel and partner OEMs and is not sold, hosted or rentable outside China, so no cloud provider on this site lists it. Huawei is on the US Entity List and Ascend silicon is treated as export-controlled wherever it is located. Huawei has also published no datasheet, product brief or catalogue entry for this card, and it does not appear in either of Huawei's public Ascend product catalogues: every figure here comes from a Huawei executive's launch remarks relayed by Chinese press.
Huawei logo

Huawei Atlas 350

Type: NPUArchitecture: Da Vinci v3Form factor: CustomReleased: 2026Spec confidence: Official
FP8 (dense)
804
TFLOPS
Memory
112 GB
HiBL 1.0
Bandwidth
1.4 TB/s
memory
TDP
600 W

Overview

The Huawei Atlas 350 is the accelerator card built on the Ascend 950PR processor, announced by Huawei VP Ma Haixu at Huawei China Partner Conference 2026 and formally on sale from 21 March 2026. It is the first Atlas card to break the long-running 300-series numbering, reflecting a new silicon generation rather than a refresh. The card is a de-rated bin of the 950PR chip rather than a straight repackaging: it carries 112 GB of HiBL 1.0 at 1.4 TB/s against the chip's 128 GB at 1.6 TB/s, exactly seven eighths of both, the signature of one memory stack fused off out of eight, and it delivers 1.56 PFLOPS of dense MXFP4 against the chip's 2 PFLOPS, a further shortfall that implies a clock or core bin on top. HiBL 1.0 is Huawei's own in-house high-bandwidth memory, developed as a lower-cost alternative to HBM3E and HBM4E and aimed at the prefill stage of inference and at recommendation workloads, which are compute-heavy and memory-light. Power is 600W, one and a half times the 400W of NVIDIA's H20, and capacity is 1.16 times the 96 GB H20, the comparison Huawei itself drew at launch. Seven Chinese server OEMs launched systems around the card, mounting up to eight of them in a 6U two-socket Kunpeng 920 chassis. Huawei has published no datasheet for the Atlas 350, no FP8 or FP16 figure, no process node and no transistor count, and its separate claim of 2.87 times the single-card compute of an H20 has no stated basis and cannot be reconciled with any published pair of numbers. The underlying silicon is catalogued separately on this site as the Ascend 950PR.

Performance

Peak theoretical throughput by precision type

PrecisionPeak
FP64
No verified data available
FP32
No verified data available
TF32
No verified data available
BF16
Brain Float 16
425TFLOPS
FP16
16-bit floating point
425TFLOPS
FP8
8-bit floating point
804TFLOPS
FP6
804TFLOPS
FP4
4-bit floating point
1,561TFLOPS
INT8
8-bit integer
804TOPS

The Atlas 350 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 MI355XAtlas 350

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

Specifications

Architecture

Da Vinci v3

Form Factor

Custom

Launch Year

2026

Process Node

No verified data available

Memory

112 GB HiBL 1.0

Bandwidth

1,400 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

Unified Bus 2.0

Spec Confidence

Official

Full Specifications

Memory
Memory 112 GB
Memory Type HiBL 1.0
Bandwidth 1.4 TB/s
Interface Width No verified data available
Interconnect & I/O
GPU-to-GPU Unified Bus 2.0
Power & Thermal
TDP 600 W
Max power (Flopper estimate) ~690 W est.
Enterprise Features
Compute APIs CANN, MindSpore, PyTorch (torch_npu), vLLM-Ascend
General
Form Factor Custom
Architecture Da Vinci v3
Process Node No verified data available
Launch Year 2026

Datasheet & Resources

Data Provenance

Every figure traced to a source

Primary Source

Publisher
Huawei
Published
No verified data available

Data Quality

Spec confidence
Official
Clock basis
Boost
Core precisions with figures
6 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 Ascend 950PR

OAM · 2026
FP32: No verified data available
Compare vs Ascend 950PR

Frequently Asked Questions

How many TFLOPS does the Huawei Atlas 350 have?

The Huawei Atlas 350 delivers 425 TFLOPS FP16 and 804 TFLOPS FP8 at peak. Flopper does not currently have a verified FP32 throughput figure for it.

What is the power consumption of the Huawei Atlas 350?

The Huawei Atlas 350 has a TDP (Thermal Design Power) rating of 600 watts.

How much memory does the Huawei Atlas 350 have?

The Huawei Atlas 350 is equipped with 112 GB of memory with 1,400 GB/s of memory bandwidth.

What architecture is the Huawei Atlas 350 based on?

The Huawei Atlas 350 is based on the Da Vinci v3 architecture, launched in 2026.

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