The CCF Computility Ecosystem Seminar 2024 will be held on August 16, 2024. With the theme of "Open Source and Open Ecology, Compilation and Parallel Optimization, and Coordinated Development of Academia and Industry", the conference aims to build a platform for exchanges and cooperation among academics, enterprises and government departments. You are cordially invited to attend this conference!
With the rapid development of artificial intelligence technology and industry, intelligent computing power has become a key force to promote the innovation of new quality productivity and industrial development. In order to deeply discuss the development opportunities and challenges of artificial intelligence computing facilities and basic system software, including compilation tool chains and common operator libraries, the first CCF High Performance Computing Committee Computing Ecological Strategy Seminar will be held in Mentougou District, Beijing on August 16, 2024.
Workshop schedule
Time: 8:00-17:30, August 16
Venue: Zhongguancun (Jingxi) Artificial Intelligence Science and Technology Park. The multi-functional hall on the first floor of the intelligent cultural and creative park
(No. 1, Qiaoyuan Road, Mentougou District, Beijing)
Workshop Schedule
August 16, 2024 | |
(Zhongguancun (Jingxi) Artificial Intelligence Science and Technology Park. Multi-functional hall on the first floor of the intelligent cultural and creative park) | |
compere | |
8:00-9:00 | Guest sign-in |
9:00-9:15 | Opening ceremony, leader's speech |
Leader of Beijing Municipal Bureau of Economy and Information Technology | |
Mentougou District Government Leader | |
9:15-9:30 | Group photo of the conference (at the entrance of the hall on the first floor) |
9:30-10:00 | Invited Report 1 - "The Road to RISC-V+AI Computing Power that Shakes the Ecological Barriers of CUDA", Tao Xie, Chair Professor, Peking University |
10:00-10:30 | Invited Lecture 2 - "Architecture and Programming System of Super Intelligent Convergence" Prof. Wenguang Chen (Department of Computer Science, Tsinghua University) |
10:30-10:45 | Tea break |
10:45-11:15 | 特邀报告3-《The Interplay among Supercomputer's Power, Resilience, and Performance》 Prof. Zizhong Chen (University of California, Riverside, United States) |
11:15-11:45 | Invited Report 4 - Intelligent Compilation and Optimization of Mathematical Software Prof. Wang Dong Yang (Hunan University) |
11:45-13:30 | lunch |
compere | |
13:30-14:00 | Invited Talk 5 - Design of Low-Latency and Scalable Cloud-Native Computing Systems Prof. Quan Chen (Department of Computer Science, Shanghai Jiao Tong University) |
14:00-14:30 | Invited Talk 6 - Data-Centric Parallel Training System for Large Models Cheng Li, Associate Professor, University of Science and Technology of China |
14:30-15:00 | Invited Report 7 - Infrastructure Design for AI Heterogeneous Platforms Bo Zhang, Chief Architect of Infrastructure Business Group (Lenovo (Beijing) Information Technology Co., Ltd.) |
15:00-15:15 | Tea break |
15:15-15:45 | Invited Talk 8 - "Maleonn: Research on Automation Performance Engineering" Wang Long, Director (Huawei Beiming Laboratory) |
15:45-16:15 | Invited Talk 9 - Stencil Parallel Algorithms and Application Optimization Liang Yuan, Associate Professor (Institute of Computing Technology, Chinese Academy of Sciences) |
16:30-17:30 | Closed-door meeting: meeting of the Standing Committee of the Director of the CCF High Commission |
Participants: Director, Deputy Director, Secretary-General and members of the Standing Committee of CCF | |
17:30 | Finish、Dinner |
(Please refer to the actual schedule of the day)
Invited speaker
Xie Tao
(Chair Professor of Peking University, Director of CCF System Software Committee, CCF Fellow, Director of the Department of Software Science and Engineering, School of Computer Science, Peking University, Chief Scientist of Beijing Open Source Chip Research Institute, Deputy Director of the Key Laboratory of High Reliability Software Technology of the Ministry of Education, Head of the National Discipline Innovation and Intelligence Introduction Base for Colleges and Universities.) )
Topic: RISC-V+AI Computing Power Shaking CUDA Ecological Barriers
Report Summary: In recent years, the RISC-V open source instruction set architecture has developed rapidly, which has become the focus of the current international scientific and technological competition, and has also become an effective starting point to build a consensus on industrial development with open source and openness, and build a global computing industry ecology. Although NVIDIA's GPUs and the CUDA software ecosystem on top of them currently dominate the global AI computing market, the industry is eager to establish a new software ecosystem to break through the barriers of the CUDA ecosystem. A consensus has gradually formed that RISC-V AI chips are commonal, uniting relevant companies and universities and research institutes to jointly formulate AI extension instruction set standards in an open source and open way, and cooperate in the development of open source AI system software stacks on them. This report discusses this direction, the major opportunities it faces and the ways in which it can address them.
Personal profile: Xie Tao is a chair professor of Peking University, director of the CCF System Software Committee, a CCF fellow, the director of the Department of Software Science and Engineering, the School of Computer Science, Peking University, the chief scientist of the Beijing Open Source Chip Research Institute, the deputy director of the Key Laboratory of High Reliability Software Technology of the Ministry of Education, and the head of the National Discipline Innovation and Intelligence Introduction Base for Colleges and Universities. He was a full professor in the Department of Computer Science at the University of Illinois at Urbana-Champaign (UIUC) in United States. He was elected as a Foreign Member of the European Academy of Sciences, a Fellow of the Association for Computing Machinery (ACM), a Fellow of the Institute of Electrical and Electronics Engineers (IEEE), a Fellow of the Association for the Advancement of Science (AAAS) in the United States, and a Fellow of the China Computer Federation (CCF). He has won the Xplorer Award, the Overseas Outstanding Young Scholar Award, the ACM China Outstanding Scholar, the National Natural Science Foundation of United States Young Career Award, and two of the three major international awards in the field of ACM Software Engineering (SIGSOFT) (Influential Educator Award, Outstanding Service Award), etc. He is the chairman of the RISC-V+AI Computing Ecology (RACE) Committee, and the chairman of the Artificial Intelligence and Machine Learning Technology Committee (AI/ML SIG) of the RISC-V International Foundation.
Chen Wenguang
(Professor, Department of Computer Science, Tsinghua University)
Topic: Architecture and Programming System of Hyper-Intelligent Convergence
Abstract: The mainland has made remarkable achievements in the field of supercomputing, but it is also facing problems such as the main investment from government funds and the main users from scientific research institutes, which poses challenges for the further development of the supercomputing field. On the other hand, the development of large artificial intelligence models has put forward a significant demand for training capabilities, and China is facing bottlenecks in intelligent computing power. From the perspective of application, in addition to large AI models, the application of AI methods in scientific computing problems also demonstrates the demand for super-intelligent integration. This report suggests that the mainland should give full play to its technological advantages in the field of supercomputing, carry out further research on architecture and programming systems, realize the integration of supercomputing, improve the efficiency of supercomputing investment, and solve the bottleneck of intelligent computing power in the mainland.
Personal Profile: Chen Wenguang is a professor in the Department of Computer Science, Tsinghua University. His main research areas are operating systems, compilers, and parallel computing. He is currently a CCF Fellow, Director of the Academic Working Committee, and an Honorary Member of YOCSEF. Vice President of Beijing Computer Federation; Executive Director of ACM China Council.
Chen Zizhong
(Professor, University of California, Riverside, United States)
报告主题:The Interplay among Supercomputer's Power, Resilience, and Performance
报告摘要:Today's supercomputers often consume megawatts level of power. While tremendous research have been done to improve applications' energy efficiency from the system point of view, few research exploits applications' algorithmic characteristics to maximize supercomputers' energy efficiency. In this talk, I will discuss our recent work on algorithmic techniques to improve applications' energy efficiency. We demonstrate that systems can not make optimal power decisions without applications' algorithmic knowledge. We investigate the interplay among application power, resilience, and performance, and demonstrate that better performance can be achieved with less power and energy when applications' algorithmic characteristics are leveraged.
个人简介:Zizhong (Jeffrey) Chen is a professor in the Department of Computer Science and Engineering at the University of California, Riverside. He is interested in high performance computing, GPU programming, distributed machine learning, and big data analytics. Collaborating with ByteDance, his group has helped to develop ByteTransformer - a GPU-based high performance Transformer that powers all of ByteDance's in house search businesses, including TikTok, Douyin, Xigua Video, Magellan, and Toutiao Search Portal, serving billions of daily active users worldwide. He has published over 150 papers with many in highly competitive conferences and journals such as SC, HPDC, ICS, SIGMOD, ICDE, PPoPP, IPDPS, TPDS, TC, TKDE, TPAMI, SIMAX, and SISC. He received a CAREER Award from the U.S. National Science Foundation and Best Paper Awards/Finalists from the IEEE International Parallel & Distributed Processing Symposium (IPDPS'23), the International Conference on Supercomputing (ICS'23), the IEEE International Conference on Cluster Computing (Cluster'18), the IEEE International Parallel & Distributed Processing Symposium (IPDPS'10) and the International Supercomputing Conference (ISC'05). Dr. Chen is a Senior Member of the IEEE and a Life Member of the ACM. He currently serves as Subject Area Editor for Elsevier Parallel Computing journal and he was Associate Editor for IEEE Transactions on Parallel and Distributed Systems from 2015-2019.
Yang Wang Dong
(Second-level Professor, Doctoral Supervisor, Hunan University, "Changjiang Scholar" of the Ministry of Education)
Topic: Intelligent compilation and optimization of mathematical software
Report Summary: With the diversification of computing power demand, processors have also diversified. On the other hand, as high-performance computing has been widely used in the social economy, the demand for the development of mathematical software is also increasing. The contradiction between the increasing development difficulty and the increasing demand urgently needs a new mathematical software development model, and a set of tool chains for intelligent compilers and optimizers for mathematical software can be established, which can not only provide compilation assistance optimization and automatic performance tuning of the source code of mathematical software for the target computing platform through the data-driven performance AI model, so as to quickly and efficiently adapt to the target computing platform and improve the computing efficiency of mathematical software.
Personal Profile: Yang Wangdong, Ph.D., is a second-level professor and doctoral supervisor of Hunan University, and a "Changjiang Scholar" of the Ministry of Education. His main research areas are high-performance computing, parallel numerical algorithms and program performance optimization, and he has long been designing and applying parallel algorithms on Tianhe series domestic supercomputing platforms and domestic processors such as Feiteng, Kunpeng, Haiguang, Ascend, and Jingjiawei GPUs, and has presided over the development of a series of basic numerical algorithm libraries, performance optimization tools and numerical simulation systems. He has won the second prize of the National Science and Technology Progress Award, the first prize and the second prize of the Natural Science Award of Hunan Province, the first prize of the China Industry-University-Research Cooperation Innovation Achievement, the special prize of the Scientific and Technological Achievements of the Chinese Computing Society, the second prize of the National Teaching Achievement Award, and the third prize of the Hunan Science and Technology Progress Award, and won two Huawei Spark Awards. He has presided over key projects and general projects of the National Natural Science Foundation of China, national projects such as national key R&D projects and core technology research projects, provincial and ministerial projects such as Hunan Provincial Key R&D Program and unveiling the leader, and scientific research projects commissioned by Huawei, Sugon and other enterprises. He has published more than 40 papers included in SCI/EI, including 12 Class A conferences and journals recommended by CCF such as SC, ICDE, TC, TPDS, etc., 13 authorized invention patents and 6 software copyrights.
Chen Quan
(Professor, Department of Computer Science, Shanghai Jiao Tong University)
Topic: Design of low-latency and scalable cloud-native computing systems
Report Summary: Cloud-native computing is the main form of next-generation cloud computing, and its main workload has the core requirements of low-latency response and efficient scaling. In view of the challenges of high competition for resources, complex operation dependencies, and lagging scaling decisions that still exist in cloud native, this report will introduce three aspects: encapsulation methods, operation mechanisms, and scaling models: 1) container encapsulation methods for the isolation of soft and hard resources, 2) efficient execution mechanisms for microservices triggered by decentralization, and 3) prior pre-scaling strategies based on the execution blocking graph model. Based on the above methods, a low-latency and high-density cloud-native runtime system "MEo" was developed to improve the efficiency of the cloud-native system. The system solution is integrated with Alibaba Cloud, Alibaba Cloud Linux, Lenovo xCloud, container cloud platform, etc., and the application results are remarkable.
Personal Profile: Chen Quan, Ph.D., is a professor and deputy director of the Department of Computer Science, Shanghai Jiao Tong University. He has been engaged in research related to computer architecture and cloud native computing for a long time. He has presided over the key projects of the National Natural Science Foundation of China, and published more than 100 academic papers in famous international conferences and journals in the fields of ASPLOS, OSDI, ATC, SC, TC, TPDS, TACO, etc. He has won the CCF Youth Science and Technology Award, Ali Green Orange Award, etc. At present, he is a member of the youth editorial board of Fundamental Research, a journal sponsored by the National Natural Science Foundation of China, and an editorial board member of the SCI journals Parallel Computing, JCST, FCS and youth editorial board members in the field. The research results have won the first prize of the 2023 CCF Technological Invention Award (ranked 1st) and the second prize of the National Technological Invention Award.
Li Cheng
(Associate Professor/Ph.D. Supervisor, University of Science and Technology of China)
Topic: Data-centric Parallel Training System for Large Models
Abstract: With the rapid development of fields such as deep learning and natural language processing, large models such as GPT-3 and its successors have become an important driving force for AI research. However, large-scale model computing requires huge computing, storage, and network resources, so it needs to be specially designed for software and hardware collaboration to meet its increasing computing power requirements. This report introduces the distributed parallel training system developed by the scientific research team of the National High Performance Computing Center (Hefei) and the Institute of Artificial Intelligence of the Hefei Comprehensive National Science Center, and proposes the collaborative optimization of memory, storage, communication and computing from the perspective of data flow, which solves the problems of "memory wall", "storage wall" and "communication wall" of large model training, and improves the efficiency of large-scale parallel computing. The system and key technologies have been applied in the pre-training and fine-tuning business of Microsoft, Trendy Technology, OPPO, Huawei, Zhongke Brain, Baidu and other companies.
Personal profile: Li Cheng, Ph.D. of the Max Planck Institute of Software Systems (MPI-SWS) of Germany, tenured associate professor and doctoral supervisor of the School of Computer Science/National High Performance Computing Center (Hefei), director of the information computing platform of the Institute of Artificial Intelligence of Hefei Comprehensive National Science Center, young teaching master and rookie in Anhui Province. Focusing on the research of large model basic system software, he has published more than 40 papers in well-known international conferences in the field of computer systems such as SOSP, OSDI, EuroSys, ATC, FAST, ASPLOS, SC, HPCA, etc. Long-term participation in the program committee of SOSP, OSDI, FAST, EuroSys, etc. He has won 10 scientific research awards, including the 2024 World Artificial Intelligence Conference Youth Outstanding Paper Award (10 papers in the world), the 2023 World Artificial Intelligence Conference Yunfan Award (10 people in the world), the 2023 Alibaba Outstanding Cooperation Project Award (15 in China), the 2022 AI 2000 Most Influential Scholar in the Field of Computer Systems nomination, the 2022 CCF Distributed Special Committee Outstanding Young Scholar, and the 2021 ACM China Rising Star nomination. The course "Compilation Principles and Techniques" was selected as the second batch of national offline first-class courses, won more than 10 teaching awards such as the first prize of the engineering group of the 5th Youth Education Competition in Anhui Province, and edited the high-quality textbooks in Anhui Province.
Zhang Bo
(Chief Architect, Infrastructure Business Group, Lenovo (Beijing) Information Technology Co., Ltd.)
Topic: AI Heterogeneous Platform Infrastructure Design
Report Summary: With the rapid development of AI computing power units, compared with traditional HPC scenarios, computing power units show a diversified development trend. At present, a large number of intelligent computing users in China have both supercomputing and intelligent computing needs, and users will encounter scheduling and management problems of heterogeneous computing power more and more frequently.
In the rapidly exploding demand for AI, heterogeneous computing power will also appear more and more in model training scenarios. How to combine dozens of hardware computing resources with hundreds of scenario requirements to maximize computing efficiency, reduce invalid computing, and improve the parallelism of model training is the main problem to be solved in the infrastructure design of AI heterogeneous platforms.
The architecture design of the AI heterogeneous computing platform can effectively solve the above problems faced by users, uniformly manage and schedule computing resources for supercomputing/intelligent computing, and coordinate the scheduling of different types of computing power units in AI scenarios with more complex computing power requirements, improve the comprehensive computing power of the system, reduce the impact of abnormal computing power units on system training, and improve training efficiency.
Personal profile: Zhang Bo, Chief Architect of Lenovo Group's Infrastructure Business Group, Chief Expert of Intelligent Computing Center Business, Senior Expert in the Meteorological and Marine Fields, has in-depth research on supercomputing/intelligent computing infrastructure design, is good at supercomputing system computing power design, has unique insights into supercomputing/intelligent computing network architecture design, high-performance storage system design, etc., and participated in the design of meteorological and marine benchmark projects such as the Beijing Winter Olympics Meteorological Forecasting System and the National Ocean High-performance Forecasting System, as well as a number of intelligent manufacturing neighborhood benchmark projects. and a number of regional supercomputing/intelligent computing center projects.
Wang Lung
(Director of Huawei Beiming Laboratory)
Topic: Maleonn: Research on Automation Performance Engineering
Report Summary: With the rise of domestic processors, the same user code often needs to develop multiple versions and spend a lot of time to optimize the performance of new processors. We believe that this obviously lengthens the iterative cycle for users to develop new algorithms, and also affects the popularization and promotion of domestic processors. How to allow users to focus more on inventing and creating application algorithms is the core driving force for our research and development of "Maleonn".
In order to achieve the goal of "automatically generating optimized code for users", we systematically defined the accelerated operators (application-level rather than traditional math libraries) in the HPC focus area, implemented the function of automatically identifying these operators from the code, and implemented efficient back-end implementation based on Huawei processors. In the report, we explain the main principles of this automation work, the accelerated results, and the future outlook.
Personal profile: Wang Long has more than ten years of research experience in high-performance computing, especially supercomputing applications, and his main work includes: 1) During the period of the Supercomputing Center of the Chinese Academy of Sciences, he presided over the research and development projects of 10000-10000-10000 cores ultra-large-scale key applications, developed a large number of high-performance computing applications, played a role in promoting the development of supercomputing applications in the mainland, and won the first prize of Beijing Science and Technology Progress Award. He is also the host of the first international conference on GPU algorithm research in China, and an early researcher and promoter of the Earth System Numerical Simulation Device. 2) During his tenure at Baidu, he served as the chairman of the technical committee of the system department, studying the intersection of compilation technology and performance optimization; 3) During his tenure at Huawei, he was in charge of the Computing System Optimization Laboratory (hereinafter referred to as Beiming), which mainly carried out performance optimization, performance portability, and software-hardware collaborative design research for HPC, AI, and general computing based on Huawei processors. He is the main inventor of the "magic pen Maleon", an automatic application optimization tool in the HPC field.
Yuan Liang
(Associate Professor, Institute of Computing Technology, Chinese Academy of Sciences)
Topic: Stencil Parallel Algorithms and Application Optimization
Abstract:Stencil is an important computing mode, which is commonly used in large-scale parallel science and engineering applications, and is also an important research object of computer parallel algorithm and compilation technology.
Biography: Liang Yuan, Institute of Computing Technology, Chinese Academy of Sciences, Associate Researcher. His main research interests are large-scale parallel application optimization, parallel algorithms and parallel computing models, and the relevant results have been published in SC, PPoPP, TPDS, TACO, ICS, IPDPS.
The organizing committee of the conference specially invites professionals who are interested in core issues such as computing power, high-performance computing, and big data, or have experience and ideas to participate. Registration is now open, please scan or identify the QR code below to register.
Scan the QR code to register
Ticket type: | Participant Identity | 7.5-8.15 |
Conference registration fee | CCF member | 1600 |
Non-Members | 2000 |
Attendance Guide
Directions
Accommodation Guide
Nearby hotels:Zhe. Coffee Hotel (Beijing Mentougou Science and Technology Park Shore Location Station Branch) is 2.3 kilometers away from the venue
Negotiated price: 438 yuan/room (brown fan big bed room) with breakfast, 478 yuan/room (brown fan view big bed room) with breakfast
Hotel Contact/Contact: Wang Meng / 13810985224
Hotel reservation method: contact the hotel contact person to make a reservation (*report computer conference, enjoy the agreed price)
Address: 1st Floor, Building 3, Lideheng Building, No. 6 Ya'an Road, Shilong Economic Development Zone, Mentougou District Tel: 010-60806410, 010-60806420
Nearby hotels: Beijing Mentougou Shang'an Subway Station MUSTEL Muwenti Hotel Distance from the venue: 2.5 kilometers
Negotiated price: 350 yuan/room (double room) with breakfast
Hotel Contact/Contact: Angela Zhang / 13681224022
Hotel reservation method: contact the hotel contact person to make a reservation (*report computer conference, enjoy the agreed price)
Hotel address: 1st Floor, Building 1, Yard 15, Caogezhuang Middle Road, Mentougou District
*Shuttle bus will be arranged
Sponsor: Lenovo Group
Homepage: https://ccf.org.cn/ccfces2024
Meeting Contact: Li Xidai 13693056420 [email protected]