NVIDIA has created a variant of the company's Hopper H100 GPU for use in China to assist with developing generative AI, such as the OpenAI ChatGPT and other AI products. This NVIDIA H800 GPU is currently used with Alibaba Group Holding, Baidu Incorporated, and Tencent Holdings.
NVIDIA confirms shipping cut-down Hopper H800 GPUs to China to limit military advancement
US regulations were implemented last year to prevent many manufacturers from importing and exporting goods to China. However, companies such as NVIDIA and Intel have found that the current regulations would need to be altered to allow for some advancements and warranty coverage for products currently in possession by companies and users.
NVIDIA confirmed today that it developed a legal version of its Hopper H100 GPU, the H800, to be solely exported to Chinese companies, albeit with much less power than its predecessor. Industry sources in China have spoken with news outlet Reuters that the H800 is reduced to half the data transfer rate of its chip-to-chip design which would involve longer training times. Whether that is the only change made to these GPUs or not remains to be seen.
The NVIDIA H800 is the second time NVIDIA has made a chip that followed the regulations set by the United States government. The first chip to have a massive cut made to its performance by the company was the NVIDIA A800, a derivative of the A100 GPU.
Last August, the US regulatory board set guidelines to limit China's ability to develop semiconductors, AI, and more, allowing the country to be unable to increase the technology used for its military. Guidelines around the use of AI technology, especially in chips, limit the "powerful computing capabilities" where data transfer speeds are crucial for machine learning models to train at superfast speeds. We will keep you updated as we learn more about the specs of the new NVIDIA Hopper H800 GPU for China.
The post NVIDIA Develops Cuts-Down Hopper H100 GPU & Offers It As H800 To China by Jason R. Wilson appeared first on Wccftech.
Refference- https://wccftech.com
0 Comments