Original title: AI chips independently developed by aridamo Institute will be 40 times more cost-effective than similar products
4hw.com.cn news on April 19: titanium media learned that Alibaba Damo Institute is developing a neural network chip -- Ali NPU, which is an AI chip and will be mainly used in image and video analysis, machine learning and other AI reasoning calculation.
In other words, the main application scenarios of this chip are intelligent vehicles, Internet of vehicles, intelligent hardware and other fields that are in hot demand for artificial intelligence. According to Alibaba, the cost performance of the chip will be 40 times that of the current similar products.
Jiao min, a researcher of aridamo Institute, said that CPU and GPU, as general-purpose computing chips, are designed to process thread logic and graphics. They have high power consumption and low cost performance when dealing with AI computing problems. In the field of AI computing, special architecture chips are urgently needed to solve the above problems.
At present, the AI chips independently developed by Alibaba are mainly used to solve the AI reasoning problems in business scenarios such as image, video recognition and cloud computing, so as to improve the operation efficiency and reduce the cost.
According to Jiao min, Ali NPU, which is independently developed by Alibaba, is based on the advantages of a large number of algorithm models accumulated by Alibaba machine intelligence technology laboratory and other teams in the AI field. According to the AI algorithm model, the micro structure and instruction set are designed to realize the largest number of AI model algorithm operations at the minimum cost.
According to the design, Alibaba's Ali NPU performance will be 10 times that of mainstream CPU and GPU architecture AI chips on the market, while the manufacturing cost and power consumption will only be half, and the cost performance ratio will exceed 40 times. In the future, the ability of Ali NPU can not only better meet the needs of video and image processing, but also output the computing power through alicloud, enabling all walks of life.
In fact, in addition to launching its own OS' yunos' (now renamed as Alios) seven years ago, Alibaba has long had a layout for talent development of its own AI chips.
According to titanium media, the 'self-developed AI chip' is a strategic component of Ali's' China Chip 'layout. At present, the chip R & D team of Damo Institute has reached dozens in the United States and Shanghai, and is expected to reach 100 by the end of the year. Previously, Ali has invested in Cambrian, barefoot networks, Shenjian, kneron, ASR, Zhongtian micro and other chip companies.
According to industry analysis, this move has a great leading effect on the independent research and development of core technology of domestic enterprises.
A few days ago, the US Department of Commerce announced that in the next seven years, its enterprises will be prohibited from selling any electronic technology or communication components to ZTE, a Chinese telecommunications equipment manufacturer.
This incident not only affected high-tech enterprises including ZTE, but also caused a profound topic in the public opinion field. The export embargo touched the pain point of the lack of core technology in China's communication industry. The problem of "lack of core and soul" has once again been placed in front of people.
On the other hand, Chinese enterprises have made unprecedented investment in science and technology research and development.
In April this year, brand finance, a British authoritative brand evaluation organization, released the "2018 list of the world's 100 most valuable technology brands", and Alibaba ranked as the most valuable technology brand in China. According to the report "2017 global innovation 1000 research" released by PwC, a research institution, Alibaba's R & D investment in 2017 reached 2.5 billion US dollars, ranking first among Chinese enterprises.