RFQ/BOM 0 Sign In / Register

Select Your Location

The domestic core breakthrough, the first fully self-developed 7nm GPGPU chip successfully "lighted up".

Published :1/18/2021 9:06:43 AM

Click Count:2103

Recently, Shanghai Tianshu Zhixin Semiconductor Co., Ltd. announced that the company's flagship 7-nanometer general-purpose parallel (GPGPU) cloud computing chip BI has been successfully "lit" recently.


Tianshu Zhixin Semiconductor pointed out that this is the first domestically-developed, truly 7-nanometer GPGPU training chip based on the GPU architecture. After mass production, it will be widely used in AI training, high-performance computing (HPC) and other scenarios to serve Education, Internet, finance, autonomous driving, medical care, security and other fields.

According to the official website, BI products were taped out in May 2020, returned in November, and successfully "lit up" in December of that year. The chip uses a 7-nanometer process and 2.5DCoWoS packaging technology, accommodates 24 billion transistors, a single core can perform 147 trillion FP16 calculations per second, and can complete artificial intelligence processing for hundreds of camera video channels per second, with performance reaching mainstream products in the market Twice.

Headquartered in Zhangjiang, Shanghai, Tianshu Zhixin is the first hardware technology company in China focusing on high-performance computing systems for GPGPU chips. Focus on the research and development of high-end general-purpose parallel computing chips at the cloud server level, aiming at the data-driven technology market represented by cloud computing, artificial intelligence, and digital transformation to solve the core computing power bottleneck problem.

According to Jiefang Daily, the full name of GPGPU is a general-purpose graphics processor, which allows the GPU, originally born for graphics and image processing, to run general-purpose computing tasks other than graphics rendering. Because of its particularly strong parallel processing capabilities and large storage bandwidth, it is favored by artificial intelligence model training and reasoning, and high-performance computing. At present, it has broad application space in the artificial intelligence market and high-performance market.