The AI chip manufacturer introduced its “Blackwell” chip family in March, with scale production set for later this year. The new processors combine two squares of silicon that are the same size as the company’s prior offering. Within the series, the B200 is 30 times faster than its predecessor at certain activities, such as serving up Chabot responses. Nvidia is partnering with Inspur, a prominent Chinese distributor, to launch and distribute the B20 chip. The innovation comes in reaction to tighter US limits on shipping advanced chips to China, which aim to curb supercomputing advancements that could boost China’s military capabilities.
In response to recent US export limitations on high-performance AI chips, Nvidia plans to create a China-specific version of their latest AI chip, the H20. This move aims to keep Nvidia’s base in the Chinese market, which was previously hindered by these restrictions.
The H20 AI chip, which is designed to comply with US export laws, will provide less computational capacity than Nvidia’s flagship H100 but is customized to satisfy the needs of the Chinese market. It is priced competitively with Huawei’s Ascend 910B processor and is projected to be available for pre-order via distributors in the second quarter of 2024. The H20 has 96GB of memory and a performance density of 2.9 TFLOPs/die, making it ideal for applications demanding fast interconnection.
This strategic move comes as Nvidia seeks to reclaim its leading market position in China, where it formerly owned over 90% of the AI chip market until US limitations were implemented. Nvidia’s creation of the H20 chip shows their commitment to meeting regulatory requirements while serving their Chinese customers.
In conclusion, Nvidia’s H20 AI chip is an important milestone in the ongoing evolution of the AI sector, balancing regulatory compliance with technological advancement. This trend emphasizes the significance of strategic adaptation in a continuously changing global environment.
