NVIDIA is developing a new China-specific version of its AI chip: HGX H20/L20 PCle/L2 PCle

According to the Science and Technology Innovation Board Daily, on November 9, industry chain officials said that NVIDIA has developed the latest improved version of AI chips specifically for the Chinese market, including HGX H20, L20 PCle and L2 PCle

Informed sources said that these three chips are all based on NVIDIA H100 improvements. NVIDIA is expected to announce them as soon as November 16, and domestic manufacturers will get samples as soon as these days.

Nvidia has not yet responded to this rumor.

Public information shows that NVIDIA H100 Tensor Core GPU adopts the new Hopper architecture, is based on TSMC N4 process, and integrates 80 billion transistors. Delivers 9x faster training for multi-expert (MoE) models compared to the previous generation.

This product is equipped with the fourth generation Tensor Core and Transformer engine (FP8 precision), and also has highly scalable NVLink interconnect technology (can connect up to 256 H100 GPUs). Compared with the HDR Quantum InfiniBand network used in the previous generation, its bandwidth is 9 times higher and the bandwidth speed is 900GB/s

When rewriting the content, the original text needs to be rewritten into Chinese, and the original sentence does not need to appear

The above is the detailed content of NVIDIA is developing a new China-specific version of its AI chip: HGX H20/L20 PCle/L2 PCle. For more information, please follow other related articles on the PHP Chinese website!