Nvidia postpones release of AI chip for Chinese market to early 2024
Nvidia is postponing the release of the H20 AI chip for the Chinese market to the first quarter of next year. According to Reuters, some server manufacturers are said to be experiencing problems integrating the AI video card into their systems.
The H20, like the H100 GPU, is based on the Hopper architecture and has 96GB of HBM3 memory. That memory makes higher speeds and sizes possible. The memory bandwidth is 4TB/s. The AI chip was introduced at the beginning of November together with the slightly less powerful L20 and L2 video cards. The three video cards have been developed specifically for the Chinese market and would not conflict with the American restrictions on AI chips to China. The release of the L20 and L2 video cards would according to Reuters are on schedule.
The US government introduced an export ban on AI video cards a year ago. As a result, Nvidia was no longer allowed to export its A100 and H100 data center GPUs to China, Russia and later also the Middle East. According to the US government, these GPUs can be used for military purposes.
Update5 p.m.: 4 Tbit/s replaced by 4 TB/s.
Nvidia H100