NVIDIA is not willing to stop selling chips to China. This is how he is dodging US sanctions

The United States wants to prevent China from showing significant progress in fields of supercomputing and military technology. The strategy? Impose trade restrictions that restrict the export of certain US technologies to the Asian country. The Department of Commerce has already set to work to achieve this, but companies are not willing to sit idly by.

NVIDIA, the multinational specialized in the development of graphics processing units, has a plan to recover part of the Chinese market lost due to trade restrictions. According to Reuters, it is about offering alternative products to those that have been banned by the United States. And it is not an easy move, because it has had to make chips specially designed for China.

NVIDIA sells A800 GPUs instead of A100

As we reported in September, NVIDIA had to stop selling two of its most advanced products to China. We are talking about the A100 and H100 GPUs, two very important pieces when developing supercomputing projects. These graphs were included in a Department of Commerce list which requires a special license to export these products.

The request for these licenses is thoroughly analyzed, but they are generally rejected, according to The New York Times. We don’t know if NVIDIA tried to get one of these licenses, but the company chose to offer China an alternative GPU known as the NVIDIA A800. The same, at the moment, is not included in the export list of the Department of Commerce.

To meet the requirements of the United States, NVIDIA had to put products with a little less power in its catalog, although that does not mean that they are no longer beastly. The A800 GPU, like the A100, integrates the 3rd generation Tensor Core cores and is 20 times more powerful in deep learning tasks than its Volta architecture predecessors, reaching 312 TFLOPs of power.

The main difference, How can we see a technical sheet, lies in the interconnection of high-speed GPUs through NVLink. That is, the technology that allows high memory scaling and performance for computers that support high workloads. In this case, the A800 has a bandwidth of 400 GB/s while the A100 reaches 600 GB/s.

Speaking to ReutersCCS Insight analyst Wayne Lam said that “The A800 appears to be a repackaged A100 GPU”. Whether this move by NVIDA will be successful in the Chinese market remains to be seen. But that is not the only challenge to overcome. You should also make sure the Commerce Department doesn’t put your new GPU on its export list any time soon.

The truth is that this situation makes something clear: the supercomputing business is too juicy to pass up. According to The Wall Street Journal, the restrictions could lose NVIDIA up to $400 million per quarter. Boosting sales of this alternative product could at least soften the economic losses.

It should be noted that the chip war is not only limited to the mentioned fields. The United States is also trying to prevent China from making its own chips. To do this, the Joe Biden administration wants prevent US citizens work on Chinese high-tech projects.

Images: Renato Ramos Puma | Nvidia

In Xataka: The next big thing in chip manufacturing is in the hands of a single company, and it’s coming in 2025