[Global Network Technology Reporter Lindy] Recently, Stanford University released the latest DAWNBench deep learning reasoning list. Domestic companies Tianshu Smart Chip, Huawei , Cambrian , and Alibaba rank in the top four on the list. Tianshu Smart Chip won the image recognition performance championship, and its performance has increased by 30% compared with the previous best institutional achievements on the list. Baidu Brain Vision Technology Team and Baidu Smart Cloud have set four world records in this list.

It is reported that Stanford University DAWNBench is jointly organized by Stanford University and world-renowned companies such as Google, Intel, Facebook, Microsoftml4, etc. It is one of the most authoritative competitions in the field of artificial intelligence. Its purpose is to promote the democratization of AI and enable AI to empower various industries in a more convenient way. It is an international authoritative benchmark test platform used to measure the end-to-end deep learning model training and reasoning performance. Its ranking reflects the current leading deep learning platform technology in the global industry. The reasoning project requires participating institutions to accurately identify and classify 50,000 pictures. The shorter the time required for each picture, the better.
reporter learned about the news from the official website of DAWNBench. Tianshu Zhixin ranked first in ImageNet inference and won the championship in image recognition performance. With an average time of 1.868ms per picture, its performance value is 30% higher than the performance of the best institutions on the previous list, and it won the latest championship of the DAWNBench deep learning reasoning list. In this ranking, Huawei, Cambrian and Alibaba ranked second, third and fourth respectively (refer to the publishing time).
In addition, Baidu Brain Vision Technology Team and Baidu Smart Cloud have set four world records in this list. The four projects are CIFAR10 training speed and cost and CIFAR10 inference speed and cost. Among them, the CIFAR training speed is 45s and the inference speed is 0.6830ms, which greatly improve the results of the original list (the original training and inference records are 74s and 0.828ms), and the cost of achieving this effect is only US$0.02 and US$0.0000002078 for Baidu Smart Cloud GPU server. Public information shows that Tianshu Intelligent, as a system technology company focusing on providing computing power products, platforms and solutions, has been focusing on the largest common divisor market for high-performance computing represented by AI since its establishment in 2015. It is an independent IP computing chip and basic computing software for artificial intelligence, and is committed to solving the most core computing power problems in the AI era.
In June this year, at the "2019 AI Hardware Summit" (Artificial Intelligence Hardware Summit) in Beijing, Li Yunpeng, founder of Tianshu Intelligence, released the ecological logic of combining software and hardware. Tianshu Intelligence started from improving software computing power, and regarded software as the key to carrying its ecosystem, and quickly launched specific application solutions in industry scenarios.