Nvidia: In an effort to stymie the growth of artificial intelligence (AI) technology in China, the US government slapped new limits on AI and HPC GPU exports to the People’s Republic this week, according to Reuters. Nvidia will be unable to sell its A800 and H800 AI and HPC GPUs to Chinese businesses as a result of the new limitations, while Biren and Moore Threads will likely lose access to sophisticated manufacturing nodes.
Last year, the United States prohibited the shipment of processors that may allow Chinese firms to build supercomputers with more than 100 FP64 PetaFLOPS or more than 200 FP32 PetaFLOPS within 41,600 cubic feet (1178 cubic meters).
Nvidia produced cut-down GPUs to sell in China
As a result, Nvidia was unable to sell its A100, A100X, and H100-series products to Chinese enterprises and was forced to design the A800 and H800 GPUs with reduced communication capabilities expressly for China in order to comply with limitations. Intel tried something similar for China with its Gaudi 2 solutions. However, while the limits hampered supercomputers and huge deployments, they had no effect on the performance of CPUs sold to Chinese enterprises.
In a filing recently, Nvidia reminded investors that the additional license requirements for shipments to China, which also apply to Saudi Arabia, the United Arab Emirates, and Vietnam, apply to the A100, A800, H100, H800, L40, L40S, and RTX 4090. The rules will be implemented in 30 days.
In addition to these limitations, the government has added Chinese GPU developers Biren Technology and Moore Threads to the Entity List, alongside Huawei. Entities providing American-developed technologies or services to firms on the list must first apply to the BIS for an export license, which is rarely granted.
According to Reuters, buried inside the 400 pages of new rules may be a lifeline for Nvidia, AMD, and Intel. BIS officials have stated that they are open to receiving suggestions from the semiconductor sector in order to continue exporting AI chips to China for small and medium-sized systems; the existing limitations are intended to limit China’s potential to create enormous supercomputers that may be used for military purposes.