On June 30th, Zhonghao Chipin introduced its latest-generation, fully independently developed high-performance TPU AI dedicated computing chip, dubbed the "Xuyu®". Concurrently, the company also rolled out an integrated hardware-software intelligent computing platform—the Taize® 2.0 AI High-Performance Intelligent Computing Platform. When compared to its first-generation counterpart, the "Shana®" chip, the new chip exhibits remarkable enhancements across various dimensions, encompassing the underlying architecture, peak computational capacity, on-chip memory, cluster interconnection, and overall computational efficiency. The single-chip mixed-precision floating-point computational prowess has soared to 896 TFLOPS, marking a threefold increase over its predecessor. Meanwhile, the 8-bit inference computational power has reached 1,792 TOPS, providing robust support for high-concurrency inference scenarios that involve massive token volumes. Notably, the rated power consumption of the single chip has been reduced by 50%, standing at 600W. The Taize® 2.0 platform, on the other hand, delivers a staggering computational power of 7.168P per machine, facilitating high-speed interconnection among 2,048 chips to cater to the distributed training needs of large models boasting trillions of parameters.