阿里平哥头自研AI芯片“真武”正式上线

Core Viewpoint - Alibaba's chip design unit, Pingtouge, has launched a high-end AI chip named "Zhenwu 810E" following reports of its potential IPO, indicating significant progress in external market penetration and product development [1][2]. Group 1: Product Launch and Features - The "Zhenwu" PPU has been deployed in multiple clusters on Alibaba Cloud, serving over 400 clients including State Grid, Chinese Academy of Sciences, Xiaopeng Motors, and Sina Weibo [1]. - The PPU features a self-developed parallel computing architecture and inter-chip communication technology, with a memory of 96G HBM2e and an inter-chip bandwidth of 700 GB/s, suitable for AI training, inference, and autonomous driving [1]. - The overall performance of the "Zhenwu" PPU surpasses Nvidia's A800 and mainstream domestic GPUs, and is comparable to Nvidia's H20, with an upgraded version reportedly outperforming Nvidia's A100 [2]. Group 2: Strategic Integration and Market Position - Alibaba is integrating its AI, cloud computing, and chip technologies through a collaboration known as "Tongyun Ge," which includes Tongyi Lab, Alibaba Cloud, and Pingtouge, allowing for focused hardware development and optimized AI frameworks [2]. - The company aims to establish "Tongyun Ge" as an AI supercomputer, leveraging its self-developed chips, leading cloud services, and open-source models to achieve maximum efficiency in training and deploying large models [2]. - Alibaba and Google are currently the only two global tech companies with competitive capabilities across large models, cloud, and chip sectors [4].