Nvidia to launch cheaper Blackwell AI chip for China amid US export restrictions – Firstpost

Nvidia will roll out a brand new synthetic intelligence chipset for China at a considerably cheaper worth than its not too long ago restricted H20 mannequin, with mass manufacturing slated to start as early as June, in response to people conversant in the scenario.

In response to two sources, the GPU, or graphics processing unit, can be included in Nvidia’s present era Blackwell-architecture AI processors and is estimated to value between $6,500 and $8,000, which is much lower than the $10,000-$12,000 the H20 bought for.

The diminished worth displays the weaker specs and simpler manufacturing wants.

It is going to be constructed on Nvidia’s RTX Professional 6000D, a server-class graphics processor, and can use customary GDDR7 reminiscence somewhat than extra refined excessive bandwidth reminiscence, in response to the 2 sources.

They added it might not use Taiwan Semiconductor Manufacturing Co’s superior Chip-on-Wafer-on-Substrate (CoWoS) packaging expertise.

The brand new chip’s worth, specs and manufacturing timing haven’t beforehand been reported.

The three sources Reuters spoke to for this text declined to be recognized as they weren’t authorised to talk to media.

An Nvidia spokesperson stated the corporate was nonetheless evaluating its “restricted” choices. “Till we decide on a brand new product design and obtain approval from the US authorities, we’re successfully foreclosed from China’s $50 billion information middle market.”

TSMC declined to remark.

Market share plunge

China stays an enormous marketplace for Nvidia, accounting for 13% of its gross sales up to now monetary yr. It’s the third time that Nvidia has needed to tailor a GPU for the world’s second-largest financial system after restrictions from US authorities who’re eager to stymie Chinese language technological growth.

After the US successfully banned the H20 in April, Nvidia initially thought of creating a downgraded model of the H20 for China, sources have stated, however that plan didn’t work out.

Nvidia CEO Jensen Huang stated final week the corporate’s older Hopper structure – which the H20 makes use of – can not accommodate additional modifications beneath present US export restrictions.

Reuters was unable to find out the product’s ultimate identify.

Chinese language brokerage GF Securities stated in a observe revealed on Tuesday that the brand new GPU would probably be known as the 6000D or the B40, although it didn’t disclose pricing or cite sources for the data.

In response to two of the sources, Nvidia can also be creating one other Blackwell-architecture chip for China that’s set to start manufacturing as early as September. Reuters was not instantly capable of verify specs of that variant.

Nvidia’s market share in China has plummeted from 95% earlier than 2022, when US export curbs that impacted its merchandise started, to 50% at the moment, Huang informed reporters in Taipei this week. Its most important competitor is Huawei which produces the Ascend 910B chip.

Huang additionally warned that if US export curbs proceed, extra Chinese language clients will purchase Huawei’s chips.

The H20 ban compelled Nvidia to write down off $5.5 billion in stock and Huang informed the Stratechery podcast on Monday that the corporate additionally needed to stroll away from $15 billion in gross sales.

The newest export restrictions launched new limits on GPU reminiscence bandwidth – a vital metric measuring information transmission speeds between the primary processor and reminiscence chips. This functionality is especially essential for AI workloads that require in depth information processing.

Funding financial institution Jefferies estimates that the brand new laws cap reminiscence bandwidth at 1.7-1.8 terabytes per second. That compares with the 4 terabytes per second that the H20 is able to.

GF Securities forecast the brand new GPU will obtain roughly 1.7 terabytes per second utilizing GDDR7 reminiscence expertise, simply inside the export management limits.