Recently, the first China-made CPU Parallel Application Challenge (CPC) hosted by the China Computer Society, the Wuxi Branch of the Chinese Computer Society, the Wuxi National Supercomputing Center, the National Supercomputing Jinan Center, and Beijing Parallel Technology Co., Ltd. Final evaluation will be successfully held in Wuxi. Attending the review committee co-chairman of the steering committee paint Feng Bin, researcher Zhang Yunquan, chairman of the judging panel, vice chairman of the review committee Professor Chen Dexun, jury members Professor Chen Yifeng, Professor Shi Xuanhua, Chen Hu Prof. Zhang Yu, Prof. Liu Xin, Researcher Fu Haohuan, Associate Professor Xue Wei, Associate Professor Huang Xiaomeng, Chen Jian, General Manager of the Organization Committee, and other experts from the industry, thank you sincerely for their leadership.
Professor Qiu Fengbin, co-chair of the CPC2017 Challenge Steering Committee
CPC chief designer, co-chair of the Steering Committee, Qiuxin Bin, researcher for the opening speech and pointed out: 'Congratulations on behalf of students from the 146 teams to stand out in the fierce competition came to Wuxi supercomputer center on-site battle duel. We held the CPC Challenge competition is designed to promote the development of domestic CPU to create a domestic industry CPU ecosystem. Wuxi ultra-Taihu Lake is the world's fastest running CPU, three consecutive World No. 1, this is China It is very significant for everyone to have the honor to enter the final and how to get in touch with the domestic CPU in order to speed up the communication between the teams so as to further promote the development of the domestic CPU and finally wish everyone a good start Score! '
CPC2017 Challenge defending experts take a group photo
CPC2017 Finals match style Morning closed with on-site optimization, the integration of the competition time and the results of the competition into the overall score among the final afternoon each team into the defense conference room alone PPT respondent, and reply to the expert's review. Competition intense competition, experts review one by one, the final selection of the winning team.
The first domestic CPU Parallel Application Challenge Finals winning team list
First Prize
Tsinghua University Untitled Diablo
'second prize'
Zhongshan University tenth orange cat
Information Engineering University Information Engineering University team 0
'Third prize'
Tsinghua University & Shandong University nuclear whisper
Shandong University drink more water to see the document
Jiuquan Satellite Launch Center Dongfeng Technology 0 team
China University of Science and Technology China University super budget goose CPC team
China Ocean University Little Tigers
'Single Award'
Innovation Excellence Award
China University of Geosciences (Wuhan) to the light of the HPC 1 group
Shanghai Jiaotong University cycling mystery team
Best new star award
China University of Geosciences (Wuhan) to light the HPC 2 groups
Chengdu University of Information Engineering coupling team
Most commercial potential award
Younger team at Qinghai University
Chengdu University of Information Technology DLLT team
CPC2017 Challenge Tournament site experts took a group photo
The summit of the Challenge2020 matchup
CPC2017 Challenge Team Responses
CPC2017 finals, led by Wu Jiming, director of Wuxi Supercomputing Center Office led the team to visit the 'light of the Lake in Taiwei Taiwei,' the students won the third with the world's No. 1 'China Shenwei Light of the Taihu Lake' (Wuxi, China) supercomputers and front-line engineers face to face close contact opportunities, and then get the quickest way to learn how to control the history of the most efficient and best performance supercomputer 'divine power · Taihu Lake light' This year is even more peak computing power 12.5 billion per second, sustained computing power of 9.3 billion times per second computing power, won the world No. 1. Supercomputer, known as the 'national heavy equipment', supercomputing is a strategic area of high technology, is competing in various countries in the world Competing for the commanding heights of science and technology is also one of the important symbols of a country's scientific and technological strength.
HPCChina2017 conference held during the same period CPC2017 awards dinner, the contest invited Qiuxiangbin researcher, Zhang Yunquan researcher, Dr. Fu Haohuan and Dr. Chen Jian for the awards dinner to make a speech .Sundan grandly for the first domestic CPU Parallel Applications Challenge The winning team presented the first prize, the second prize, the third prize and the single award, and the scene was unprecedentedly shocked. The successful holding of the CPC Challenge was a grand revitalization of China and the promotion of China's prestige. I wish the domestic CPU parallel application challenge will be better and better , Thriving.
Special guest speaker
Awards photo
After a fierce competition in the preliminary round, a total of 16 teams entered the final stage of the CPC finals of the question is parallel implementation of the FFT algorithm for a full range of performance optimization.FFT is a DFT efficient algorithm called Fast Fourier Transform (Fast Fourier Transform), which is based on the discrete, Fourier transform odd, even, imaginary, real and other characteristics of the discrete Fourier transform algorithm to improve the FFT algorithm in the field of scientific computing has a wide range of applications.
The final team conducted an in-depth study on the FFT algorithm, and combined with the heterogeneous architecture of the Shenwei 26010 chip used in the computing system of "Shenwei • Taihu Lake Light", a number of optimization methods were designed to greatly improve the program's performance in ' Taiwanese Light 'computing system computing efficiency compared to the original version of the code, the team gained up to 150 times the speedup; relative to the Intel INTEL Xeon multi-core platform, the results obtained a maximum of 120 times the acceleration ratio.
CPC2017 Challenge team entries highlight the technical highlights
• Extensive program performance analysis using multiple platform tools to quickly find program performance bottlenecks and implement appropriate optimizations.
• Loading computational data into efficient acceleration calculations from the core array leverages the power of Shenwei CPUs by designing and implementing optimization approaches that closely integrate with the SW26010 processor architecture.
• Utilizes the register communication mechanism unique to the SW26010 chip to enable fast data exchange between cores.
• Utilize a highly efficient matrix transpose approach from the core.
• Design efficient inter-process communication schemes to hide communication and computation time.
• Use the SIMD interface provided by the Divine platform to improve parallel computing efficiency.
Manually rearrange the assembly code to achieve instruction flow.
• Merge adjacent transpose operations and FFT calculations to reduce the number of slave DMAs.
• The "Shenwei • Taihu Lake Light" computing system is the result of major national 863 project research and is the first supercomputer built in China with a domestic processor and developed by the National Center for Parallel Computing and Engineering Technology. In 2016 6 The TOP500 supercomputer rankings on the 20th of 20th were the average of the three key indicators of 'peak power of Shenwei • Taihu Lake' system (125.436PFlops), continuous operation performance (93.015PFlops) and performance power ratio (6.05GFlops / W) Habitat in the world.
'Shenwei · Taihu Lake Light' computing system contains a total of 40,960 'Shen Wei 26010' all nuclear processors. 'Shen Wei 26010' is the country's 'nuclear high base' major projects supported by China's first independent research and development of the nuclear The processor, developed by the National High Performance Integrated Circuit Design Center, has led the world in performance and successfully mass-produced, breaking the technical blockade imposed by the United States on our country. The processor is based on the SW-64 instruction set and adopts on-chip fusion heterogeneous Core architecture and FCBGA3832 package, a single processor contains 260 computing cores.
'Divine · Taihu Lake Light' has the world's leading ultra-large-scale system of low-power control technology and high-density assembly, saving more than 60% energy than the current world's second-ranked system, single machine bin packing density ranks first in the world.At the same time, 'Divine · Taihu Lake Light' system independent research and development software to establish a high-performance computing software based on Shen Wei CPU ecological chain.
Recently, the first China-made CPU Parallel Application Challenge (CPC) hosted by the China Computer Society, the Wuxi Branch of the Chinese Computer Society, the Wuxi National Supercomputing Center, the National Supercomputing Jinan Center, and Beijing Parallel Technology Co., Ltd. Final evaluation will be successfully held in Wuxi. Attending the review committee co-chairman of the steering committee paint Feng Bin, researcher Zhang Yunquan, chairman of the judging panel, vice chairman of the review committee Professor Chen Dexun, jury members Professor Chen Yifeng, Professor Shi Xuanhua, Chen Hu Prof. Zhang Yu, Prof. Liu Xin, Researcher Fu Haohuan, Associate Professor Xue Wei, Associate Professor Huang Xiaomeng, Chen Jian, General Manager of the Organization Committee, and other experts from the industry, thank you sincerely for their leadership.
Professor Qiu Fengbin, co-chair of the CPC2017 Challenge Steering Committee
CPC chief designer, co-chair of the Steering Committee, Qiuxin Bin, researcher for the opening speech and pointed out: 'Congratulations on behalf of students from the 146 teams to stand out in the fierce competition came to Wuxi supercomputer center on-site battle duel. We held the CPC Challenge competition is designed to promote the development of domestic CPU to create a domestic industry CPU ecosystem. Wuxi ultra-Taihu Lake is the world's fastest running CPU, three consecutive World No. 1, this is China It is very significant for everyone to have the honor to enter the final and how to get in touch with the domestic CPU in order to speed up the communication between the teams so as to further promote the development of the domestic CPU and finally wish everyone a good start Score! '
CPC2017 Challenge defending experts take a group photo
CPC2017 Finals match style Morning closed with on-site optimization, the integration of the competition time and the results of the competition into the overall score among the final afternoon each team into the defense conference room alone PPT respondent, and reply to the expert's review. Competition intense competition, experts review one by one, the final selection of the winning team.
The first domestic CPU Parallel Application Challenge Finals winning team list
First Prize
Tsinghua University Untitled Diablo
'second prize'
Zhongshan University tenth orange cat
Information Engineering University Information Engineering University team 0
'Third prize'
Tsinghua University & Shandong University nuclear whisper
Shandong University drink more water to see the document
Jiuquan Satellite Launch Center Dongfeng Technology 0 team
China University of Science and Technology China University super budget goose CPC team
China Ocean University Little Tigers
'Single Award'
Innovation Excellence Award
China University of Geosciences (Wuhan) to the light of the HPC 1 group
Shanghai Jiaotong University cycling mystery team
Best new star award
China University of Geosciences (Wuhan) to light the HPC 2 groups
Chengdu University of Information Engineering coupling team
Most commercial potential award
Younger team at Qinghai University
Chengdu University of Information Technology DLLT team
CPC2017 Challenge Tournament site experts took a group photo
The summit of the Challenge2020 matchup
CPC2017 Challenge Team Responses
CPC2017 finals, led by Wu Jiming, director of Wuxi Supercomputing Center Office led the team to visit the 'Light of Taihu Lake Taiwei', the students won the third with the world's No. 1 'China Shenwei Light of the Taihu Lake' (Wuxi, China) supercomputers and front-line engineers face to face close contact opportunities, and then get the quickest way to learn how to control the history of the most efficient and best performance supercomputer 'divine power · Taihu Lake light' This year is even more peak computing power 12.5 billion per second, sustained computing power of 9.3 billion times per second computing power, won the world No. 1. Supercomputer, known as the 'national heavy equipment', supercomputing is a strategic area of high technology, is competing in various countries in the world Competing for the commanding heights of science and technology is also one of the important symbols of a country's scientific and technological strength.
HPCChina2017 conference held during the same period CPC2017 awards dinner, the contest invited Qiuxiangbin researcher, Zhang Yunquan researcher, Dr. Fu Haohuan and Dr. Chen Jian for the awards dinner to make a speech .Sundan grandly for the first domestic CPU Parallel Applications Challenge The winning team presented the first prize, the second prize, the third prize and the single award, and the scene was unprecedentedly shocked. The successful holding of the CPC Challenge was a grand revitalization of China and the promotion of China's prestige. I wish the domestic CPU parallel application challenge will be better and better , Thriving.
Special guest speaker
Awards photo
After a fierce competition in the preliminary round, a total of 16 teams entered the final stage of the CPC finals of the question is parallel implementation of the FFT algorithm for a full range of performance optimization.FFT is a DFT efficient algorithm called Fast Fourier Transform (Fast Fourier Transform), which is based on the discrete, Fourier transform odd, even, imaginary, real and other characteristics of the discrete Fourier transform algorithm to improve the FFT algorithm in the field of scientific computing has a wide range of applications.
The final team conducted an in-depth study on the FFT algorithm, and combined with the heterogeneous architecture of the Shenwei 26010 chip used in the computing system of "Shenwei • Taihu Lake Light", a number of optimization methods were designed to greatly improve the program's performance in ' Taiwanese Light 'computing system computing efficiency compared to the original version of the code, the team gained up to 150 times the speedup; relative to the Intel INTEL Xeon multi-core platform, the results obtained a maximum of 120 times the acceleration ratio.
CPC2017 Challenge team entries highlight the technical highlights
• Extensive program performance analysis using multiple platform tools to quickly find program performance bottlenecks and implement appropriate optimizations.
• Loading computational data into efficient acceleration calculations from the core array leverages the power of Shenwei CPUs by designing and implementing optimization approaches that closely integrate with the SW26010 processor architecture.
• Utilizes the register communication mechanism unique to the SW26010 chip to enable fast data exchange between cores.
• Utilize a highly efficient matrix transpose approach from the core.
• Design efficient inter-process communication schemes to hide communication and computation time.
• Use the SIMD interface provided by the Divine platform to improve parallel computing efficiency.
Manually rearrange the assembly code to achieve instruction flow.
• Merge adjacent transpose operations and FFT calculations to reduce the number of slave DMAs.
• The "Shenwei • Taihu Lake Light" computing system is the result of major national 863 project research and is the first supercomputer built in China with a domestic processor and developed by the National Center for Parallel Computing and Engineering Technology. In 2016 6 The TOP500 supercomputer rankings on the 20th of 20th were the average of the three key indicators of 'peak power of Shenwei • Taihu Lake' system (125.436PFlops), continuous operation performance (93.015PFlops) and performance power ratio (6.05GFlops / W) Habitat in the world.
'Shenwei · Taihu Lake Light' computing system contains a total of 40,960 'Shen Wei 26010' all nuclear processors. 'Shen Wei 26010' is the country's 'nuclear high base' major projects supported by China's first independent research and development of the nuclear The processor, developed by the National High Performance Integrated Circuit Design Center, has led the world in performance and successfully mass-produced, breaking the technical blockade imposed by the United States on our country. The processor is based on the SW-64 instruction set and adopts on-chip fusion heterogeneous Core architecture and FCBGA3832 package, a single processor contains 260 computing cores.
'Divine · Taihu Lake Light' has the world's leading ultra-large-scale system of low-power control technology and high-density assembly, saving more than 60% more than the current world's second-ranked system, single unit compact packaging density ranks first in the world. 'Divine · Taihu Lake Light' system independent research and development software to establish a high-performance computing software based on Shen Wei CPU ecological chain.
