Chinese think tank recommends Nvidia chips for data centres, warning of high transfer costs
The Beijing-based think tank warned that transferring AI models trained on Nvidia GPUs to domestic solutions involved ‘complex engineering’
A government-backed think tank has suggested that data centres in mainland China should choose Nvidia chips, warning of the high costs involved in shifting to domestic solutions.
“If the conditions allow, [data centres] can choose [Nvidia’s] A100 and H100 high-performance computing units. If the need for computing power is limited, they can also choose H20 or alternative domestic solutions,” the China Academy of Information and Communications Technology (CAICT) said in a report on China’s computing power development issued on Sunday.
Nvidia’s chips have been highly coveted in China by big internet companies and other entities alike, but the California-based company is facing intensified US pressure not to ship its advanced graphics processing units (GPUs) to the mainland, its third-largest market in the financial year ended January 28, over fears the technology will be used to enhance the Chinese military.
Amid US sanctions, China’s domestic GPU start-ups have flourished, with progress in both hardware and software over the past three years, according to CAICT. However, the think tank pointed out that “it involves complex engineering to transfer models trained on Nvidia GPUs to domestic solutions due to differences in hardware and software”.
The A100 and H100 data centre GPUs – two of the most in-demand chips for training and running AI models – have been barred from export to China since August 2022. To get around the restrictions, Nvidia modified those chips to create the A800 and H800, but Washington subsequently barred them from being sold to China last October.
