Shulou(Shulou.com)11/24 Report--

CTOnews.com June 1 news, Nvidia CEO Huang Renxun said after his keynote speech at the Taipei computer Show that he would strive to diversify the supply chain and said he was open to working with Intel to carry artificial intelligence chips in the future. As a result, some people suspect that Intel will make top chips such as the H100 for Nvidia.

Huang Jen-hsun now says that the H100 is produced exclusively by TSMC and will not consider adding a second wafer foundry. The main reason is that "it is very difficult to do it once, and it is more difficult to do it twice separately from two foundries."

Nvidia and TSMC have begun 3/2nm cooperation planning, and Huang Jen-hsun also confirmed for the first time that TSMC will continue to "gobble up big orders" for the next generation of AI chips, at least in the AI GPU series.

It is worth mentioning that Nvidia A100 / A800 and H100 have both issued orders for TSMC, the RTX 40 series has also fully returned to TSMC 4nm (CTOnews.com Note: the RTX 30 series uses Samsung 8nm process), the current cooperation between Nvidia and Samsung is mainly memory.

Nvidia H100 GPU uses TSMC 4nm process, A100 is 7nm process, while Nvidia's top H100 Tensor Core GPU adopts Hopper architecture, based on TSMC 4nm process, it is equipped with fourth generation Tensor Core and Transformer engine, single card is equipped with 80GB memory, which is twice as efficient as A100 GPU per kilowatt.

For now, TSMC does not see a real competitor in supporting the sub-7nm process of AI chip manufacturing. Intel has returned to foundry, but has not made any meaningful progress.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.