Huawei to Transform Chip Game

For three decades, NVIDIA has topped the charts in the GPU space. Intel, AMD and Apple, all formidable companies, have been trying for decades but NVIDIA remains unbeatable, indispensable. The world’s biggest AI company, OpenAI, and almost all other AI labs, for that matter, use the chip giant’s GPUs to train their AI models.

The wind blew in favour of NVIDIA — until now when it witnessed direct competition from the Chinese conglomerate Huawei. Its Ascend 910 AI processor, which is based on Ascend-Max chipsets, has GPU capabilities on par with NVIDIA’s A100. Liu Qingfeng, founder and chairman of iFlytek, who is building a general-purpose AI model to compete with OpenAI’s ChatGPT, is using Ascend 910 for model training. 

Interestingly, the Ascend 910 accelerator provides 256 TeraFLOPs of power for tensor floating-point operations, consuming only 310 W of maximum power. In comparison, the NVIDIA A100 delivers 312 TeraFLOPs but requires 400 W of maximum power.