Huawei Large Model performance exceeds 1.1 times of mainstream GPUs

A few days ago, Huawei Cloud Global Marketing and Sales Services, Shi Jilin said that its large model training with Pangu 3.0 has reached 1.1 times more powerful than the mainstream GPUs in the industry. However, the Huawei official didn’t name any GPU for specific.

Jilin said Huawei’s AIGC technology based on large models has been applied in many categories including movies and TVs, games, audio, animation, and more.

Looking at the AI computing power, Huawei Cloud brings industry customers Ascend AI cloud services, with a single-cluster 2000P Flops surging computing power.

In the inference scenario, after the original model is migrated to the Ascend AI cloud service. Its AI inference performance is improved by 30% compared with the mainstream in the industry.

In the large model training scenario, the training performance of Huawei Pangu tech has reached 1.1 times that of the mainstream GPUs in the industry.

The Huawei Cloud marketing chief mentioned that Huawei’s Cloud can provide customers with full-stack AI services. The Pangu Large Model 3.0 includes a three-layer architecture, which fully meets the needs of industry application large models.

(via)