Nvidia tweaks flagship H100 chip for export to China as H800

Nvidia Corp., the U.S. semiconductor designer that dominates the marketplace for synthetic intelligence (AI) chips, stated it has modified its flagship product right into a model that’s authorized to export to China.

U.S. regulators final yr put into place guidelines that stopped Nvidia from promoting its two most superior chips, the A100 and newer H100, to Chinese language clients citing nationwide safety considerations. Such chips are essential to growing generative AI applied sciences like OpenAI’s ChatGPT and comparable merchandise.

Reuters in November reported that Nvidia had designed a chip referred to as the A800 that diminished some capabilities of the A100 to make the A800 authorized for export to China.

On Tuesday, the corporate stated it has equally developed a China-export model of its H100 chip. The brand new chip, referred to as the H800, is being utilized by the cloud computing items of Chinese language expertise corporations comparable to Alibaba Group Holding Ltd., Baidu Inc. and Tencent Holdings Ltd., an organization spokesperson stated.

(For high expertise information of the day, subscribe to our tech e-newsletter At this time’s Cache)

“For the younger startups, a lot of them who’re constructing giant language fashions now, and lots of of them leaping onto the generative AI revolution, they’ll sit up for Alibaba, Tencent and Baidu to have glorious cloud capabilities with Nvidia’s AI,” Nvidia Chief Government Jensen Huang informed reporters on Wednesday morning Asia time.

U.S. regulators final fall imposed guidelines to sluggish China’s growth in key expertise sectors comparable to semiconductors and synthetic intelligence, aiming to hobble the nation’s efforts to modernise its army.

The principles round AI chips imposed a take a look at that bans these with each highly effective computing capabilities and excessive chip-to-chip knowledge switch charges. Switch velocity is essential when coaching synthetic intelligence fashions on large quantities of information as a result of slower switch charges imply extra coaching time.

A chip business supply in China informed Reuters the H800 primarily diminished the chip-to-chip knowledge switch price to about half the speed of the flagship H100.

The Nvidia spokesperson declined to say how the China-focused H800 differs from the H100, besides that “our 800 sequence merchandise are absolutely compliant with export management rules.”

That is your final free article.

Supply hyperlink

Leave a Reply

Your email address will not be published. Required fields are marked *