Select Language:
Alibaba’s chip division, T-Head Semiconductor, has announced details about its self-developed AI chip, Zhenwu 810E, amid rumors of a potential IPO. The chip features 96GB of High Bandwidth Memory 2e and offers an inter-chip bandwidth of up to 700 GB per second. It is designed for various applications such as multimodal model training, inference, and autonomous driving.
The Zhenwu 810E outperforms Nvidia’s A800 and is comparable to Nvidia’s H20, both of which have been tailored for the Chinese market with reduced performance to adhere to US export restrictions on high-end chips.
On January 22, Bloomberg reported that Alibaba intends to support T-Head’s independent listing, marking the company’s move in a broader trend of Chinese chipmakers going public, alongside firms like Moore Threads Technology, Metax Integrated Circuits, and Biren Intelligent Technology. Attempts to obtain comment from Alibaba about this were unsuccessful.
According to sources in Hangzhou, Alibaba has already integrated the Zhenwu 810E into the training and inference processes of its own large language model, Qianwen. The chip, combined with Alibaba’s cloud services, is currently serving over 400 clients, including State Grid, the Chinese Academy of Sciences, and electric vehicle manufacturer Xpeng.
Founded in 2018, T-Head was formed through the merger of Alibaba’s Damo Academy chip team and C-Sky Microsystems, a chip design company acquired by Alibaba. The division has launched several products, such as the Hanguang 800 AI inference chip, the Yitian 710 server CPU, and the Zhenyue 510 SSD controller.
– End of content.
ChatGPT
Perplexity
Gemini AI
Grok AI
