A government-backed think tank has suggested that mainland Chinese data centers should choose Nvidia chips, warning that moving to domestic solutions would be costly.
“If conditions permit, (data centers) can choose (Nvidia’s) A100 and H100 high-performance computing units. If computing power needs are limited, choose the H20 or an alternative domestic solution. ,” the China Academy of Information and Communication Technology (CAICT), a government-backed think tank, said in a report on China’s computing capacity development released in June 2019. Sunday.
Nvidia’s chips are highly coveted in China by major internet companies and other companies alike, but the California-based company is the third largest manufacturer of advanced graphics processing units (GPUs) in China. are facing increasing pressure from the United States not to ship to the mainland, which is a big market for them. The fiscal year ended Jan. 28 amid concerns that the technology could be used to strengthen China’s military.
Amid US sanctions, China’s domestic GPU startups have flourished, making advances in both hardware and software over the past three years, CAICT said. However, the think tank noted that “transitioning models trained on Nvidia GPUs to domestic solutions will require complex engineering due to differences in hardware and software.”
A100 and H100 data center GPUs, two of the most sought-after chips for training and running AI models, will be banned from export to China starting August 2022. To get around the limitations, Nvidia modified these chips to create the A800 and H800. But then Washington banned sales to China last October.
