China-based AI startup Enflame Technology has released Cloud Blazer, its second-generation AI inference acceleration card equipped with in-house-developed AI inference chip Suixi 2.5, according to company founder and COO Arthur Zhang.
Based on second-generation GCU-CARA architecture, Suixi 2.5 is produced using GlobalFoundries 12nm FinFET process, with computing capability comparable with that of 7nm GPUs.
Other China-based startups have also scored points in developing AI chips. YITU Technology in 2019 unveiled QuestCore, an AI deep learning SoC for application to cloud computing then; and Horizon Robotics in 2019 developed an AIoT chip, Sunrise 2, through integrating second-generation Brain Processing Unit (BPU) architecture, with the chip reaching 4TOPS.
Besides startups, China-based high-tech giants have been actively developing AI chips. T-Head Semiconductor, Alibaba Group's wholly owned subsidiary, unveiled Hanguang 800 AI inference chip in 2019. Hanguang 800 was used to support Taobao.com's mobile search and personalized recommendation services at Double 11 online shopping festival in 2021.
Starting production of first-generation Kunlun AI chip in early 2020, Baidu began production of second-generation Kunlun based on in-house-developed XPU architecture using 7nm process in August 2021, with the chip to be applied to autonomous driving, smart transportation and AI assistant.
Tencent in November 2021 unveiled three AI chips: Zixiao inference chip, Changhai video transcoding chip, network interface controller Xuanling.
Huawei Technologies unveiled Ascend 310 AI inference chip in 2018 and Ascend 910 AI training chip in 2019.
Article translated by Adam Hwang
