Sources say China's special version of Nvidia H20 AI chip has been postponed to the first quarter of next year

Chip consulting company SemiAnalysis previously reported that Nvidia has developed the latest improved version of AI chips for China, including HGX H20, L20 PCIe and L2 PCIe, which are more than 20% faster than H100 in LLM inference.

According to Reuters, two sources revealed that Nvidia has informed Chinese customers that the launch of the HGX H20 AI chip will be postponed to the first quarter of next year.

According to sources, they received notification that the release of the H20 was delayed because the server manufacturer encountered problems integrating the chips. Some of them said they were told it might be released in February or March

Both people familiar with the matter declined to be named as the information is confidential. Nvidia declined to comment

According to sources, in addition to the H20 chip, the L20 chip will not be delayed and will be launched as originally planned. They cannot reveal information about the status of the L2 chip

IT House previously reported that Semianalysis published a paid article that NVIDIA will soon launch HGX H20, L20 PCle and L2 PCle interface chips. These three chips are all based on NVIDIA H100 improvements. Officials have produced these GPU samples and pre-released them. It is estimated to be put into mass production next month.

"Special statement: The content of the above works (including videos, pictures or audios) are users of "Dafenghao", a self-media platform owned by ifeng.com

Source: IT Home

The above is the detailed content of Sources say China's special version of Nvidia H20 AI chip has been postponed to the first quarter of next year. For more information, please follow other related articles on the PHP Chinese website!