Liu Qingfeng, founder and chairman of Chinese AI company iFlytek, stated at the 19th Summer Summit of the 2023 Yabuli China Entrepreneurs Forum that Huawei’s compute GPU capabilities are now comparable to Nvidia’s A100 GPUs (via IT Home).
According to Liu Qingfeng, Huawei has advanced significantly in the GPU industry, gaining capabilities and performance on par with Nvidia’s A100 GPU. If accurate, this would be a tremendous achievement given Nvidia’s illustrious history of superiority in AI and high-performance computing.
Along with underlining the strategic significance of this technological achievement for Huawei and the larger Chinese AI ecosystem, Liu also mentioned that three board members from Huawei are working closely with iFlytek’s special team.
iFlytek’s statements are effectively the first indication that such a device exists because Huawei has never acknowledged the creation of its own compute GPUs.
Ascend 910 AI accelerators are actually being utilized by the business for Huawei’s Atlas 900 Pod A2 AI training cluster, which Huawei has been using for its AI initiatives for a while. But, the iFlytek CEO made it clear that he was speaking about a computational GPU, which suggests a new product.
Nvidia’s Superior Computational Power
Notwithstanding the strength of Chinese AI algorithms, Liu stated that Nvidia has always outperformed domestic technology in terms of computational power. He mentioned the difficulties Chinese businesses have had while training massive AI models, which is typically done on Nvidia hardware.
Any reliance on Nvidia’s compute GPUs is a key barrier for Chinese AI companies since access to Nvidia’s hardware is somewhat restricted for Chinese enterprises due to U.S. curbs against the country’s supercomputer sector.
Furthermore disclosed by Liu were iFlytek’s aspirational goals to compete with ChatGPT and ultimately GPT-4 by the first half of 2024 by releasing a general-purpose AI model by October.
A cognitive large-scale model with seven basic competencies—text production, language understanding, knowledge-based question-answering, logical reasoning, mathematical ability, coding ability, and multi-modal capability—was formally introduced by iFlytek in May.
The model is expected to achieve milestones in open-ended question answering and multi-modal interactions, making it a versatile tool in the field of artificial intelligence.