H20 successor in China will not be hopper based, says NVIDIA CEO CEO

Nvidia will have to give up its top -of -the -line architecture to try to offer accelerators to China, according to the company’s own CEO. Quoting news from TV Taiwanesa, Reuters reports that Jensen Huang, founder and head of Nvidia, spoke about the subject recently, making it clear that the company will not give up the important Chinese market.

It is interesting to remember a little about the trajectory of export ban on GeForce’s products to the Chinese market. The H100 and H200 models, main hopper -based products, had their export prevented rapidly. Thus, Nvidia created the H800 in March 2023, to meet the rules of the sanctions imposed by the US.

Shortly thereafter, these models were also prevented from exporting, giving rise to the H20. These accelerators, then, are even more limited versions of the hopper chip, but they were still successful in China.

Now that the H20 models have also been barred by the US government, Nvidia does not intend to do another variant based on the same generation. “It will not be hopper, because it is impossible to modify the hopper even more,” said Huang.

Nvidia can give up using hbm too

US authorities mentioned the H20 memory band as part of their official justification for the impediment of exports. Thus, it is speculated that Nvidia can abandon HBM in the next products aimed at China.

This aligns with Huang’s recent speech, as Hopper architecture was fully designed for use in conjunction with HBM memories. Using another technology, it would be easier to give up the high band memories in favor of GDDR7.

Related News:

Thus, Nvidia would be among the most current Blackwell, which is focused on home GPUS and less advanced professional use, or Ada Lovelace, past generation of its most advanced accelerators.

Ada may end up being the final option due to its nvlink support, crucial for the use of several GPUs on scale.

Via: Tom’s Hardware