ã€World Wide Web Reporter Zhang Zhiying】 “China is one of the countries that have invested heavily in HPC and AI. I am very proud to be in China and China is very innovative.†Senior Director of High Performance Computing and Artificial Intelligence at Lenovo's Data Center Business Group Scott Tease said. He worked at the U.S. headquarters and traveled to China last week to participate in the 2017 National High Performance Computing Academic Annual Conference (HPC China 2017). Scott Tease, Senior Director, High Performance Computing and Artificial Intelligence Technology, Lenovo's Data Center Business Group In recent years, Lenovo has made remarkable achievements in the field of super-computing, occupying 91 seats in the TOP 500 supercomputer ranking, equivalent to one in every 5 supercomputers. Not only deployment areas include China and overseas; industry applications range from the energy industry, petroleum, exploration, chip design companies, nuclear energy technology, communications, and a series of universities and colleges at Peking University. The most recent achievement, Lenovo and the Chinese Academy of Sciences Institute of Mathematics and Systems Science have created a new high-performance cluster project. The cluster has adopted Lenovo's HPC overall construction plan and has 408 computing nodes, achieving 3,080,000 per second far exceeding customer expectations. Millions of computing speeds. The cooperation between the two parties also has a small story. According to the disclosure, when the Chinese Academy of Sciences chose a partner, a very complicated mathematics question was issued. As a result, Lenovo scored the highest score. Lenovo's performance in supercomputers is well-founded. Under the hot wave of artificial intelligence, the influence of AI spreads to the field of super-computing, including, of course, Lenovo. How is Lenovo doing artificial intelligence? One example is that they have developed a software that specifically searches for social media content. This software captures images, videos, and text that contain associative content, and then comprehensively analyzes the information to see if the user’s evaluation and sentiment about Lenovo is positive or not. Negative. In other words, through AI machine learning, the analysis of these data will be transmitted to the product department, which can become an important reference for product refinement. Scott Tease said that the current increase in work value through artificial intelligence is more and more common in Lenovo. Artificial intelligence also brings different ideas to the hypercomputing field because supercomputing has always been dedicated to finding precise answers. “We have some ways to ensure 100% correct answers. For example, math problems will have the correct answer. If you are in banking or chemical analysis, a precise answer is a must,†Scott Tease continued, “but In the AI ​​field, we do not need 100% accurate answers. The speed of processing is more important than accuracy." Scott Tease pointed out that in the AI ​​field, you will see a low-precision approach, as long as the answer is similar. Since it is never possible to obtain a 100% accurate answer in AI, the goal of AI is to obtain an approximate answer as quickly as possible. According to Scott Tease, AI learning models need to run over and over again. The more times they run and the faster they get, the better the AI ​​effect can be. For AI, this low-precision, high-speed processing is very important, it will change the way of programming, and change the data center architecture, mode. Since HPC is different from AI, how can supercomputing be used in artificial intelligence? In this regard, Scott Tease stated that AI needs huge amounts of data. With massive amounts of data, it is necessary to store them first. The next step is to train these machines to digest the data, obtain machine learning capabilities through massive data analysis, and find trends. One can imagine that high-performance networks, accelerated computing performance, open source software, which HPC is good at, are suitable for AI, can also be converted into a solid foundation for AI development. AI technology is very dependent on big data. The deployment of data determines how to train machines for learning. If the data is in the cloud, this kind of machine training must also happen in the cloud. If the data is not on the cloud but on site, it is difficult to have enough network capacity. All the massive data is transmitted to the cloud, so machine learning must also be on site. In short, where the data is, it determines where the machine learning occurs, so the solution is not just a cloud solution. "We expect there will be many different types of processors entering the market, including new Intel processors, new NVIDIA processors, TPUs, FPGAs and other innovative technologies will come into being, and our programming model will also change greatly." Scott Tease predicts. Although artificial intelligence seems to be an inevitable trend, Scott Tease believes that the three major obstacles to the rapid deployment of AI in enterprises are: first, how to structure the data; second, there are too few good AI consultants to understand how to The deployment of AI technology to different industries is even more scarce. Third, in the AI ​​world, many software is open source, but customers are also skeptical about deploying open source software from the Internet to deploy in their data centers. “Some commercial customers may not be HPC customers. They are just commercial customers. They are accustomed to buying software from SAP or Microsoft, and they can buy it, but in the AI ​​field, many of the software is open source. The industry also needs to have enough AI software and Hardware combined resources."