Nvidia tweaks flagship H100 chip for export to China as H800
© Reuters. FILE PHOTO: H100, Nvidia’s latest GPU optimized to handle large artificial intelligence models used to create text, computer code, images, video or audio is seen in this photo.” Santa Clara, CA U.S.,September 2022. NVIDIA/Handout via REUTERS
By Stephen Nellis and Jane Lanhee Lee
SAN FRANCISCO (Reuters) -Nvidia Corp, the U.S. semiconductor designer that dominates the market for artificial intelligence (AI) chips, said it has modified its flagship product into a version that is legal to export to China.
U.S. regulators last year put into place rules that stopped Nvidia (NASDAQ:) from selling its two most advanced chips, the A100 and newer H100, to Chinese customers citing national security concerns. Such chips are crucial to developing generative AI technologies like OpenAI’s ChatGPT and similar products.
Reuters in November reported that Nvidia had designed a chip called the A800 that reduced some capabilities of the A100 to make the A800 legal for export to China.
On Tuesday, the company said it has similarly developed a China-export version of its H100 chip. The new chip, called the H800, is being used by the cloud computing units of Chinese technology firms such as Alibaba (NYSE:) Group Holding Ltd, Baidu Inc (NASDAQ:) and Tencent Holdings (OTC:) Ltd, a company spokesperson said.
“For the young startups, many of them who are building large language models now, and many of them jumping onto the generative AI revolution, they can look forward to Alibaba, Tencent and Baidu to have excellent cloud capabilities with Nvidia’s AI,” Nvidia Chief Executive Jensen Huang told reporters on Wednesday morning Asia time.
U.S. regulators last fall imposed rules to slow China’s development in key technology sectors such as semiconductors and artificial intelligence, aiming to hobble the country’s efforts to modernize its military.
The rules around AI chips imposed a test that bans those with both powerful computing capabilities and high chip-to-chip data transfer rates. Transfer speed is important when training artificial intelligence models on huge amounts of data because slower transfer rates mean more training time.
A chip industry source in China told Reuters the H800 mainly reduced the chip-to-chip data transfer rate to about half the rate of the flagship H100.
The Nvidia spokesperson declined to say how the China-focused H800 differs from the H100, except that “our 800 series products are fully compliant with export control regulations.”