►National Super Computing Wuxi Center. Photography: Di Lihui
Written by | Di Lihui (The chief editor of "Intellectual")
Editor | Li Xiaoming
Intellectuals for a better intellectual life ID: The-Intellectual
● ● ●
June 25, 2018 International Super held in Frankfurt, Germany At the conference, the latest TOP500 global supercomputer ranking will be officially released soon. The supercomputer Summit of Oak Ridge National Laboratory in the United States will be listed at a peak calculation performance of 2 billion times per second (200 PFlops) Top the world's fastest supercomputer.
China's Tianhe No. 2 and Shenwei Taihu Light gave up the first place that has occupied 5 years. However, in the opinion of industry insiders, Summit has been postponed by one year from the expected date, and China is not unable to build 200PFlops machines, but each has its own established development rhythm. It is important to compete for the world's strongest computer, but the real competition for supercomputers is not just about who is the first.
1. Use
Although many years have passed, Fu Haohuan still remembers such a story. It was in November 2010, when Tianhe 1A became the fastest computer in the world for the first time. Experts in the Chinese supercomputer industry are very happy. After all, the honor of "fastest" has been monopolized by the United States and Japan for decades, and other countries have never been involved in it.
However, at the US Supercomputer Conference where Tianhe 1A won the world's first championship, an American expert asked, "China now has the ability to build the world's fastest supercomputer, but can it use the fastest machine to solve the top ones." Science or engineering issues? In the eyes of this expert, this behemoth may be "better but not useful".
Although the expert's criticism is sharp and harsh, it does reflect a fact: it is good to build the fastest computer, and whether it can truly work is the foundation.
At about that time, Fu Haohuan returned to Tsinghua University from Stanford and joined the recently established Department of Earth Systems Science. It was also from the colleagues who came back from the meeting that Fu Haohuan heard about this story. At that time, Tianhe 1A used Intel processors and Nvidia's GPU accelerator, and it was just starting to run large applications on it.
"The second year after I returned to Tsinghua University, in 2011, it happened that several of us slowly got to know each other during that year. I first met Teacher Xue Wei, and then after being introduced by Teacher Xue, I met again. Yang Chao teacher. "Fu Haohuan said.
Professor Xue Wei from the Department of Computer Science at Tsinghua University was previously engaged in power grid research, and then expanded to other application fields such as climate simulation. Researcher Yang Chao from the Institute of Software, Chinese Academy of Sciences is engaged in the research of computational mathematics. Fu Haohuan was fascinated by the new computing architecture at that time. "Because Tianhe 1A is a CPU plus GPU, the main computing power is provided by the GPU accelerator. So to use it efficiently, we have to use the GPU well. We formed a team at that time and wanted to use this machine. Solve the problem of atmospheric simulation." Fu Haohuan told Intellectuals.
Starting from Tianhe 1A, in just a few years, China has launched a series of supercomputers, constantly breaking the "fastest" record. In October 2011, the "Shenwei Blu-ray" that settled in Jinan all uses domestic processors and system software and hardware, with peak calculations exceeding tens of trillions of times. In June 2013, the "Tianhe 2" developed by the National University of Defense Science and Technology became the "fastest in the world" with a peak calculation speed of 549 million times per second. In June 2016, the "Shenwei Taihu Light" developed by the National Parallel Computer Engineering Technology Research Center had a peak calculation speed of 1.25 billion times per second, and its performance was more stable. Since June 2013, supercomputers from China have occupied the first place on the TOP500 list for five years.
In the past six years, it has not only changed the increasingly powerful machines.
► At the 2017 International High Performance Computing Conference held at Denver , the nonlinear large earthquake simulation application based on Shenwei Taihu Light won the highest award in the field of supercomputing application: the Gordon Bell Award. Image provided by the respondent.
In November 2016, at the International Supercomputing Conference held in Salt Lake City, USA, a research team composed of Yang Chao, Xue Wei, Fu Haohuan and others, together with Beijing Normal University, won the "Gordon Bell" Award. This highest award in the field of high-performance computing applications has been monopolized by supercomputing powers such as the United States and Japan for decades. Not only did China break this record, it also occupied 3 seats in the 6 shortlisted applications. In November 2017, the cross-border team led by Fu Haohuan won the award again. Fu Haohuan is currently an associate professor in the Department of Earth Systems Sciences at Tsinghua University and deputy director of the National Supercomputing Wuxi Center.
Looking back suddenly, the pattern of world supercomputing has changed.
In 2016, in a report on high-performance computing, The US National Security Agency and the Department of Energy believe that China's supercomputer has been close to the United States (near-peer) for its national security, economy, society, and supercomputer Industry and other industries pose threats. If they do not increase investment and seek change, the United States will lose its leadership.
Nowadays, no one will think that Chinese supercomputers are just behemoths. "The Light of Taihu Lake is an important progress for China. Indeed, its speed of 93Pflops per second is higher than the total of all computers of the US Department of Energy; more importantly, China's supercomputing systems were not only running standards (such as There are not many large-scale applications outside the LINPACK test), but this time, the large-scale programs on Taihu Light can run well." The report said.
Nowadays, the strongest competition in the field of supercomputing has become a competition between a few countries such as China, the United States, and Japan. On June 25, Summit from the Oak Ridge National Laboratory was officially released, returning to the first place. However, the industry knows that who has the best machine is just one step in the supercomputer competition.
► Shenwei·The Light of Taihu Lake. Photography: Di Lihui
2. Create a complete ecosystem
Using the powerful computing power of Taihu Light, Fu Haohuan's current work is hoped to provide scientists with a more accurate simulation platform for studying the mechanism of earthquake occurrence, while also providing earthquake resistance and earthquake resistance. Provide quantitative evaluation tools. Since winning the "Gordon Bell" award in simulating the Tangshan earthquake last year, his team aimed at the next more difficult goal, simulating Wenchuan earthquake .
On May 12, ten years ago, an earthquake in Wenchuan, Sichuan caused nearly 70,000 deaths, more than 370,000 injuries, and more than 170,000 missing, shocking the world. However, unlike forecasting weather, facing the deep crust structure that cannot be directly detected, numerical forecasting of earthquakes is still an unsolvable problem. To settle for the second best, Fu's team plans to use supercomputers to simulate the seismic source generation and seismic wave propagation process of earthquakes, complete earthquake damage prediction based on earthquake process scenario simulation for specific areas, and quantify the evaluation of earthquake intensity and earthquake damage distribution. Accurate simulation can play a significant role in reducing and preventing huge losses caused by earthquake disasters, as well as carrying out reasonable planning and design of earthquake-prone areas.
During the Wenchuan earthquake, a large number of landslides buried villages and towns, forming a landslide lake, endangering the lives of thousands of people. Fu's team plans to work with the team studying landslides at Chengdu University of Technology to study whether the earthquake will trigger a landslide and what kind of landslides will trigger. In addition, they also worked with building earthquake prevention experts to connect the model of the earthquake with the model of the building to study what damage it will cause to the building after an earthquake. "Easy of the , California, area in the United States, there are frequent earthquakes, where people have to buy earthquake insurance when buying a house. The pricing of insurance for each house depends on quantitative risk assessment based on earthquake simulation." Fu Haohuan introduced It is said that earthquake-related industrial services cannot be separated from accurate simulation.
From earthquake simulation, atmospheric simulation, ocean simulation, to titanium alloy microstructure simulation, Tokmak escape current simulation, and three-dimensional electronic structure calculation of protein cryo-electron microscopy, scientists are using supercomputers to solve important scientific problems. In addition to scientific users, various supercomputing platforms have also accepted users of many companies - they want to design new aircraft engines, ships, and turbine blades for wind power generation.Supercomputing has become an indispensable and important part of supporting intelligent manufacturing and improving economic quality. The hope behind the large-scale investment of the
supercomputing platform is that it is greater socio-economic benefits. Taking Shenwei Taihu Light as an example, in addition to the central government's investment of 1.8 billion yuan, Jiangsu Province and Wuxi City, each invested 600 million yuan. In addition, Wuxi City is also responsible for daily operating expenses, including bearing 15 MW of power consumption (approximately equivalent to the power consumption of 3 Tsinghua University). Wuxi City hopes to attract enterprises and researchers engaged in artificial intelligence, intelligent manufacturing, and industrial big data through supercomputing platforms. The supercomputing industrial park it has established also hopes to benefit local economic development.
As the director of the National Supercomputing Wuxi Center, Yang Guangwen takes every opportunity to promote the light of Shenwei Taihu Lake. At a forum in June last year, after introducing Shenwei, he earnestly told the audience that he hoped that everyone could use the major equipment of this country, solve important scientific problems, and promote industrial upgrading. "I want to say that supercomputers sound high-end and national strategy, but will they make it a sustainable (industry)? Supercomputers also want to succeed," he said.
In the aforementioned report by the NSA and the Department of Energy, it also believes that the development of Chinese supercomputing will threaten the U.S. economy: "The loss of high-performance computing leadership potentially threatens industries that have already used high-performance computing heavily. US companies. These industries include automobiles, aerospace, advanced manufacturing, petroleum exploration, pharmaceutical research, finance, etc. These industries are already highly competitive globally, so some damage to US competitiveness (for example, if China uses it High-performance computing leadership provides cheap or even free computing power to its domestic industries) will translate into greater economic effects. It will also bring significant intellectual property risks: if a pharmaceutical company's drug development is underway, then To what extent will IP leakage hinder the competitiveness of the United States? "
. For China, the development of some industries (such as the design of aircraft engines) has reached a certain level, and software must be developed by itself, and it is impossible to do it like in the past. Use foreign commercial software directly. "Including aircraft or high-speed rail, we mainly rely on foreign commercial software for design and simulation. In the next step, we hope to build on Shenwei's platform, especially with these application units, to cooperate in depth to do our independent industrial simulation and design. Software platform.” Fu Haohuan said.
From using domestic processors to compilers, operating systems, various computing libraries, and development frameworks, Chinese supercomputers are trying to build their own ecosystem, and to truly enter a healthy development, it still depends on the use of users and Feedback, they are a crucial part of the supercomputer community.
3. Looking forward to the entire Qinghai-Tibet Plateau
On April 9, 2015, the US Department of Commerce issued an announcement banning the sale of Intel "Xiang" chips to four national supercomputing centers in China. Some outside comments believe that this embargo may be related to Tianhe 2's multiple occupancy of the top supercomputer high-performance rankings. Immediately afterwards, at the end of 2015, China launched the Shenwei Taihu Light, replacing Tianhe 2, continuing to become the fastest computer, and using nationally produced chips.
However, these two things are just a "coincidence" and there is no direct connection between them. The previous "Shenwei Blu-ray" has already adopted national chips. Shenwei Taihu Light has been produced nationwide since its design, but it was launched according to the expected progress.
This is exactly the launch of Summit in the United States (actually delayed by one year than expected), and it is expected for the supercomputing industry. "Can we create 200P and 300P? In fact, it can be created, but it just means that everyone's planning rhythm is different." Fu Haohuan explained.
At present, countries around the world are preparing to build the next generation of E-class computers (10 billion times) to push computing power to a new level.Regarding why E-class computers are built, Professor Jack Dongarra, the TOP500 sponsor, Oak Ridge National Laboratory and University of Tennessee, told Intellectuals during the ASC conference held in Nanchang last month: "It's not that we want to build the next generation of computers," It is scientists who have raised more difficult challenges and more difficult problems (such computers are needed). E-level computers are just an obstacle that we have crossed in the development, and there will be higher-level computers in the future."
"Knowledge" "Molecule" learned that there are currently three institutions in China that are independently developing E-class computers and are expected to be launched in 2020. For China's supercomputers, a more important and longer-term task may be the construction of the entire ecosystem.
"From supercomputing alone, China has already achieved breakthrough results that are comparable to the United States and Japan, whether it is hardware systems or software applications in some fields. But it is necessary to be able to form a positive perspective on the many industries behind supercomputing. Feedback will take a long time. For example, weather and climate prediction based on supercomputing, industrial design and simulation, preparation of new materials, new drug research and development, etc., any industry can be integrated with supercomputing hardware and software. Huge industrial improvement. We now have a breakthrough in supercomputing technology, and we should look forward to how to achieve radiation and support to the industry in the next step." Fu Haohuan said.
For China's supercomputer, now there is a Mount Everest, and the next step should be to look forward to the entire Qinghai-Tibet Plateau.
References:
1. U.S. Leadership in High Performance Computing (HPC): A Report from the NSA-DOE Technical Meeting on High Performer ce Computing, December 1, 2016.
2. Haohuan Fu, Conghui He, Bingwei Chen, Zekun Yin, Zhengu Zhang, Wenqiang Zhang, Tingjian Zhang, Wei Xue, Weiguo Liu, Wanwang Yin, Guangwen Yang, Xiaofei Chen: 18.9-Pflops Nonlinear Earthquake Simulation on Sunway Ta ihuLight: Enabling Depiction of 18-Hz and 8-Meter Scenarios, the International Conference for High Performance Computing, Networking, Storage and Analysis (SC 2017), Denver, USA, November, 2017. (ACM Gordon Bell Prize)
Platformed Editor: Scarlett|
This page has not published content Reproduction and use of
is prohibited with written permission Please contact the authorization for reprinting of official accounts, newspapers, etc.
Business cooperation please contact
Intellectuals for better intellectual life ID: The-Intel lentual
