Huawei Atlas 350 vs Huawei Ascend 950PR
The Ascend 950PR delivers 24% more FP8 throughput than the Atlas 350 (1,000 vs 804 TFLOPS dense). The Ascend 950PR also carries 16 GB more memory (128 GB vs 112 GB).
Huawei Atlas 350
Full specs →Huawei Ascend 950PR
Full specs →Specifications
Performance (TFLOPS)
FLOPS by Precision
What actually differs
Both GPUs launched in 2026: the Atlas 350 on Da Vinci v3 and the Ascend 950PR on Da Vinci v3.
Dense throughput for the Atlas 350 against the Ascend 950PR: FP16 425 vs 500 TFLOPS, FP8 804 vs 1,000 TFLOPS. Sparse figures, where the vendor publishes them, appear under each dense number in the performance table.
Memory is 112 GB against 128 GB, fed at 1.4 TB/s versus 1.6 TB/s. More memory per GPU means larger models fit before you have to shard across cards, which often matters more than raw TFLOPS for inference.
For LLM training and inference, weight the FP8 and FP16 rows and memory capacity most heavily. For scientific and HPC workloads, the FP64 row is the one to read.
Frequently asked questions
Is the Atlas 350 faster than the Ascend 950PR?
At FP8 precision the Ascend 950PR reaches 1,000 TFLOPS dense against 804 TFLOPS for the Atlas 350. The performance table on this page lists every published precision for both GPUs.
Which has more memory, the Atlas 350 or the Ascend 950PR?
The Ascend 950PR carries 128 GB of memory versus 112 GB for the Atlas 350. Memory bandwidth is 1.4 TB/s for the Atlas 350 and 1.6 TB/s for the Ascend 950PR.
How much power do the Atlas 350 and the Ascend 950PR draw?
The Atlas 350 is rated at 600 W TDP and the Ascend 950PR at 900 W.
Get Comparison Updates
New GPUs added weekly. Be the first to see how they compare.