Exascale Computing Study: Technology Challenges in Achieving Exascale Systems

Total Page:16

File Type:pdf, Size:1020Kb

Exascale Computing Study: Technology Challenges in Achieving Exascale Systems ExaScale Computing Study: Technology Challenges in Achieving Exascale Systems Peter Kogge, Editor & Study Lead Keren Bergman Shekhar Borkar Dan Campbell William Carlson William Dally Monty Denneau Paul Franzon William Harrod Kerry Hill Jon Hiller Sherman Karp Stephen Keckler Dean Klein Robert Lucas Mark Richards Al Scarpelli Steven Scott Allan Snavely Thomas Sterling R. Stanley Williams Katherine Yelick September 28, 2008 This work was sponsored by DARPA IPTO in the ExaScale Computing Study with Dr. William Harrod as Program Manager; AFRL contract number FA8650-07-C-7724. This report is published in the interest of scientific and technical information exchange and its publication does not constitute the Government’s approval or disapproval of its ideas or findings NOTICE Using Government drawings, specifications, or other data included in this document for any purpose other than Government procurement does not in any way obligate the U.S. Government. The fact that the Government formulated or supplied the drawings, specifications, or other data does not license the holder or any other person or corporation; or convey any rights or permission to manufacture, use, or sell any patented invention that may relate to them. APPROVED FOR PUBLIC RELEASE, DISTRIBUTION UNLIMITED. This page intentionally left blank. DISCLAIMER The following disclaimer was signed by all members of the Exascale Study Group (listed below): I agree that the material in this document reects the collective views, ideas, opinions and ¯ndings of the study participants only, and not those of any of the universities, corporations, or other institutions with which they are a±liated. Furthermore, the material in this document does not reect the o±cial views, ideas, opinions and/or ¯ndings of DARPA, the Department of Defense, or of the United States government. Keren Bergman Shekhar Borkar Dan Campbell William Carlson William Dally Monty Denneau Paul Franzon William Harrod Kerry Hill Jon Hiller Sherman Karp Stephen Keckler Dean Klein Peter Kogge Robert Lucas Mark Richards Al Scarpelli Steven Scott Allan Snavely Thomas Sterling R. Stanley Williams Katherine Yelick i This page intentionally left blank. ii FOREWORD This document reects the thoughts of a group of highly talented individuals from universities, industry, and research labs on what might be the challenges in advancing computing by a thousand- fold by 2015. The work was sponsored by DARPA IPTO with Dr. William Harrod as Program Manager, under AFRL contract #FA8650-07-C-7724. The report itself was drawn from the results of a series of meetings over the second half of 2007, and as such reects a snapshot in time. The goal of the study was to assay the state of the art, and not to either propose a potential system or prepare and propose a detailed roadmap for its development. Further, the report itself was assembled in just a few months at the beginning of 2008 from input by the participants. As such, all inconsistencies reect either areas where there really are signi¯cant open research questions, or misunderstandings by the editor. There was, however, virtually complete agreement about the key challenges that surfaced from the study, and the potential value that solving them may have towards advancing the ¯eld of high performance computing. I am honored to have been part of this study, and wish to thank the study members for their passion for the subject, and for contributing far more of their precious time than they expected. Peter M. Kogge, Editor and Study Lead University of Notre Dame May 1, 2008. iii This page intentionally left blank. iv Contents 1 Executive Overview 1 2 De¯ning an Exascale System 5 2.1 Attributes . 5 2.1.1 Functional Metrics . 5 2.1.2 Physical Attributes . 6 2.1.3 Balanced Designs . 6 2.1.4 Application Performance . 7 2.2 Classes of Exascale Systems . 8 2.2.1 Data Center System . 8 2.2.2 Exascale and HPC . 9 2.2.3 Departmental Systems . 9 2.2.4 Embedded Systems . 10 2.2.5 Cross-class Applications . 11 2.3 Systems Classes and Matching Attributes . 12 2.3.1 Capacity Data Center-sized Exa Systems . 12 2.3.2 Capability Data Center-sized Exa Systems . 13 2.3.3 Departmental Peta Systems . 14 2.3.4 Embedded Tera Systems . 14 2.4 Prioritizing the Attributes . 14 3 Background 17 3.1 Prehistory . 17 3.2 Trends . 18 3.3 Overall Observations . 19 3.4 This Study . 19 3.5 Target Timeframes and Tipping Points . 20 3.6 Companion Studies . 20 3.7 Prior Relevant Studies . 21 3.7.1 1999 PITAC Report to the President . 21 3.7.2 2000 DSB Report on DoD Supercomputing Needs . 21 3.7.3 2001 Survey of National Security HPC Architectural Requirements . 21 3.7.4 2001 DoD R&D Agenda For High Productivity Computing Systems . 22 3.7.5 2002 HPC for the National Security Community . 22 3.7.6 2003 Jason Study on Requirements for ASCI . 23 3.7.7 2003 Roadmap for the Revitalization of High-End Computing . 23 3.7.8 2004 Getting Up to Speed: The Future of Supercomputing . 24 v 3.7.9 2005 Revitalizing Computer Architecture Research . 24 3.7.10 2006 DSB Task Force on Defense Critical Technologies . 25 3.7.11 2006 The Landscape of Parallel Computing Research . 25 4 Computing as We Know It 27 4.1 Today's Architectures and Execution Models . 27 4.1.1 Today's Microarchitectural Trends . 27 4.1.1.1 Conventional Microprocessors . 28 4.1.1.2 Graphics Processors . 28 4.1.1.3 Multi-core Microprocessors . 28 4.1.2 Today's Memory Systems . 29 4.1.3 Unconventional Architectures . 30 4.1.4 Data Center/Supercomputing Systems . 31 4.1.4.1 Data Center Architectures . 31 4.1.4.2 Data Center Power . 32 4.1.4.2.1 Mitigation . 33 4.1.4.3 Other Data Center Challenges . 33 4.1.5 Departmental Systems . 34 4.1.6 Embedded Systems . 34 4.1.7 Summary of the State of the Art . 35 4.2 Today's Operating Environments . 35 4.2.1 Unix . 36 4.2.2 Windows NT Kernel . 37 4.2.3 Microkernels . 37 4.2.4 Middleware . 38 4.2.5 Summary of the State of the Art . 38 4.3 Today's Programming Models . 38 4.3.1 Automatic Parallelization . 40 4.3.2 Data Parallel Languages . 40 4.3.3 Shared Memory . 41 4.3.3.1 OpenMP . 42 4.3.3.2 Threads . 43 4.3.4 Message Passing . 44 4.3.5 PGAS Languages . 45 4.3.6 The HPCS Languages . 46 4.4 Today's Microprocessors . 47 4.4.1 Basic Technology Parameters . 47 4.4.2 Overall Chip Parameters . 49 4.4.3 Summary of the State of the Art . 53 4.5 Today's Top 500 Supercomputers . 53 4.5.1 Aggregate Performance . 53 4.5.2 E±ciency . 54 4.5.3 Performance Components . 54 4.5.3.1 Processor Parallelism . 55 4.5.3.2 Clock . 56 4.5.3.3 Thread Level Concurrency . 56 4.5.3.4 Total Concurrency . 57 4.5.4 Main Memory Capacity . 59 vi 5 Exascale Application Characteristics 61 5.1 Kiviat Diagrams . 61 5.2 Balance and the von Neumann Bottleneck . 62 5.3 A Typical Application . 63 5.4 Exascale Application Characteristics . 65 5.5 Memory Intensive Applications of Today . 66 5.5.1 Latency-Sensitive Applications . 66 5.5.2 Locality Sensitive Applications . 68 5.5.3 Communication Costs - Bisection Bandwidth . 69 5.6 Exascale Applications Scaling . 71 5.6.1 Application Categories . 71 5.6.2 Memory Requirements . 72 5.6.3 Increasing Non-Main Memory Storage Capacity . 73 5.6.3.1 Scratch Storage . ..
Recommended publications
  • Materials Modelling and the Challenges of Petascale and Exascale
    Multiscale Materials Modelling on High Performance Computer Architectures Materials modelling and the challenges of petascale and exascale Andrew Emerson Cineca Supercomputing Centre, Bologna, Italy The project MMM@HPC is funded by the 7th Framework Programme of the European Commission within the Research Infrastructures 26/09/2013 with grant agreement number RI-261594. Contents Introduction to HPC HPC and the MMM@HPC project Petascale computing The Road to Exascale Observations 26/09/2013 A. Emerson, International Forum on Multiscale Materials Modelling, Bologna 2013 2 High Performance Computing High Performance Computing (HPC). What is it ? High-performance computing (HPC) is the use of parallel processing for running advanced application programs efficiently, reliably and quickly. The term applies especially to systems that function above a teraflop or 10 12 floating- point operations per second. (http://searchenterpriselinux.techtarget.com/definition/high -performance -computing ) A branch of computer science that concentrates on developing supercomputers and software to run on supercomputers. A main area of this discipline is developing parallel processing algorithms and software: programs that can be divided into little pieces so that each piece can be executed simultaneously by separate processors. (WEBOPEDIA ) 26/09/2013 A. Emerson, International Forum on Multiscale Materials Modelling, Bologna 2013 3 High Performance Computing Advances due to HPC, e.g. Molecular dynamics early 1990s . Lysozyme, 40k atoms 2006. Satellite tobacco mosaic virus (STMV). 1M atoms, 50ns 2008. Ribosome. 3.2M atoms, 230ns. 2011 . Chromatophore, 100M atoms (SC 2011) 26/09/2013 A. Emerson, International Forum on Multiscale Materials Modelling, Bologna 2013 4 High Performance Computing Cray-1 Supercomputer (1976) 80MHz , Vector processor → 250Mflops Cray XMP (1982) 2 CPUs+vectors, 400 MFlops “FERMI”, Bluegene/Q 168,000 cores 2.1 Pflops 26/09/2013 A.
    [Show full text]
  • 2020 ALCF Science Report
    ARGONNE LEADERSHIP 2020 COMPUTING FACILITY Science On the cover: A snapshot of a visualization of the SARS-CoV-2 viral envelope comprising 305 million atoms. A multi-institutional research team used multiple supercomputing resources, including the ALCF’s Theta system, to optimize codes in preparation for large-scale simulations of the SARS-CoV-2 spike protein that were recognized with the ACM Gordon Bell Special Prize for HPC-Based COVID-19 Research. Image: Rommie Amaro, Lorenzo Casalino, Abigail Dommer, and Zied Gaieb, University of California San Diego 2020 SCIENCE CONTENTS 03 Message from ALCF Leadership 04 Argonne Leadership Computing Facility 10 Advancing Science with HPC 06 About ALCF 12 ALCF Resources Contribute to Fight Against COVID-19 07 ALCF Team 16 Edge Services Propel Data-Driven Science 08 ALCF Computing Resources 18 Preparing for Science in the Exascale Era 26 Science 28 Accessing ALCF GPCNeT: Designing a Benchmark 43 Materials Science 51 Physics Resources for Science Suite for Inducing and Measuring Constructing and Navigating Hadronic Light-by-Light Scattering Contention in HPC Networks Polymorphic Landscapes of and Vacuum Polarization Sudheer Chunduri Molecular Crystals Contributions to the Muon 30 2020 Science Highlights Parallel Relational Algebra for Alexandre Tkatchenko Anomalous Magnetic Moment Thomas Blum 31 Biological Sciences Logical Inferencing at Scale Data-Driven Materials Sidharth Kumar Scalable Reinforcement-Learning- Discovery for Optoelectronic The Last Journey Based Neural Architecture Search Applications
    [Show full text]
  • Seminar HPC Trends Winter Term 2017/2018 New Operating System Concepts for High Performance Computing
    Seminar HPC Trends Winter Term 2017/2018 New Operating System Concepts for High Performance Computing Fabian Dreer Ludwig-Maximilians Universit¨atM¨unchen [email protected] January 2018 Abstract 1 The Impact of System Noise When using a traditional operating system kernel When running large-scale applications on clusters, in high performance computing applications, the the noise generated by the operating system can cache and interrupt system are under heavy load by greatly impact the overall performance. In order to e.g. system services for housekeeping tasks which is minimize overhead, new concepts for HPC OSs are also referred to as noise. The performance of the needed as a response to increasing complexity while application is notably reduced by this noise. still considering existing API compatibility. Even small delays from cache misses or interrupts can affect the overall performance of a large scale In this paper we study the design concepts of het- application. So called jitter even influences collec- erogeneous kernels using the example of mOS and tive communication regarding the synchronization, the approach of library operating systems by ex- which can either absorb or propagate the noise. ploring the architecture of Exokernel. We sum- Even though asynchronous communication has a marize architectural decisions, present a similar much higher probability to absorb the noise, it is project in each case, Interface for Heterogeneous not completely unaffected. Collective operations Kernels and Unikernels respectively, and show suffer the most from propagation of jitter especially benchmark results where possible. when implemented linearly. But it is hard to anal- yse noise and its propagation for collective oper- ations even for simple algorithms.
    [Show full text]
  • The Linux Device File-System
    The Linux Device File-System Richard Gooch EMC Corporation [email protected] Abstract 1 Introduction All Unix systems provide access to hardware via de- vice drivers. These drivers need to provide entry points for user-space applications and system tools to access the hardware. Following the \everything is a file” philosophy of Unix, these entry points are ex- posed in the file name-space, and are called \device The Device File-System (devfs) provides a power- special files” or \device nodes". ful new device management mechanism for Linux. Unlike other existing and proposed device manage- This paper discusses how these device nodes are cre- ment schemes, it is powerful, flexible, scalable and ated and managed in conventional Unix systems and efficient. the limitations this scheme imposes. An alternative mechanism is then presented. It is an alternative to conventional disc-based char- acter and block special devices. Kernel device drivers can register devices by name rather than de- vice numbers, and these device entries will appear in the file-system automatically. 1.1 Device numbers Devfs provides an immediate benefit to system ad- ministrators, as it implements a device naming scheme which is more convenient for large systems Conventional Unix systems have the concept of a (providing a topology-based name-space) and small \device number". Each instance of a driver and systems (via a device-class based name-space) alike. hardware component is assigned a unique device number. Within the kernel, this device number is Device driver authors can benefit from devfs by used to refer to the hardware and driver instance.
    [Show full text]
  • Biology at the Exascale
    Biology at the Exascale Advances in computational hardware and algorithms that have transformed areas of physics and engineering have recently brought similar benefits to biology and biomedical research. Contributors: Laura Wolf and Dr. Gail W. Pieper, Argonne National Laboratory Biological sciences are undergoing a revolution. High‐performance computing has accelerated the transition from hypothesis‐driven to design‐driven research at all scales, and computational simulation of biological systems is now driving the direction of biological experimentation and the generation of insights. As recently as ten years ago, success in predicting how proteins assume their intricate three‐dimensional forms was considered highly unlikely if there was no related protein of known structure. For those proteins whose sequence resembles a protein of known structure, the three‐dimensional structure of the known protein can be used as a “template” to deduce the unknown protein structure. At the time, about 60 percent of protein sequences arising from the genome sequencing projects had no homologs of known structure. In 2001, Rosetta, a computational technique developed by Dr. David Baker and colleagues at the Howard Hughes Medical Institute, successfully predicted the three‐dimensional structure of a folded protein from its linear sequence of amino acids. (Baker now develops tools to enable researchers to test new protein scaffolds, examine additional structural hypothesis regarding determinants of binding, and ultimately design proteins that tightly bind endogenous cellular proteins.) Two years later, a thirteen‐year project to sequence the human genome was declared a success, making available to scientists worldwide the billions of letters of DNA to conduct postgenomic research, including annotating the human genome.
    [Show full text]
  • Recent Developments in Supercomputing
    John von Neumann Institute for Computing Recent Developments in Supercomputing Th. Lippert published in NIC Symposium 2008, G. M¨unster, D. Wolf, M. Kremer (Editors), John von Neumann Institute for Computing, J¨ulich, NIC Series, Vol. 39, ISBN 978-3-9810843-5-1, pp. 1-8, 2008. c 2008 by John von Neumann Institute for Computing Permission to make digital or hard copies of portions of this work for personal or classroom use is granted provided that the copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise requires prior specific permission by the publisher mentioned above. http://www.fz-juelich.de/nic-series/volume39 Recent Developments in Supercomputing Thomas Lippert J¨ulich Supercomputing Centre, Forschungszentrum J¨ulich 52425 J¨ulich, Germany E-mail: [email protected] Status and recent developments in the field of supercomputing on the European and German level as well as at the Forschungszentrum J¨ulich are presented. Encouraged by the ESFRI committee, the European PRACE Initiative is going to create a world-leading European tier-0 supercomputer infrastructure. In Germany, the BMBF formed the Gauss Centre for Supercom- puting, the largest national association for supercomputing in Europe. The Gauss Centre is the German partner in PRACE. With its new Blue Gene/P system, the J¨ulich supercomputing centre has realized its vision of a dual system complex and is heading for Petaflop/s already in 2009. In the framework of the JuRoPA-Project, in cooperation with German and European industrial partners, the JSC will build a next generation general purpose system with very good price-performance ratio and energy efficiency.
    [Show full text]
  • Multi-Core Processors and Systems: State-Of-The-Art and Study of Performance Increase
    Multi-Core Processors and Systems: State-of-the-Art and Study of Performance Increase Abhilash Goyal Computer Science Department San Jose State University San Jose, CA 95192 408-924-1000 [email protected] ABSTRACT speedup. Some tasks are easily divided into parts that can be To achieve the large processing power, we are moving towards processed in parallel. In those scenarios, speed up will most likely Parallel Processing. In the simple words, parallel processing can follow “common trajectory” as shown in Figure 2. If an be defined as using two or more processors (cores, computers) in application has little or no inherent parallelism, then little or no combination to solve a single problem. To achieve the good speedup will be achieved and because of overhead, speed up may results by parallel processing, in the industry many multi-core follow as show by “occasional trajectory” in Figure 2. processors has been designed and fabricated. In this class-project paper, the overview of the state-of-the-art of the multi-core processors designed by several companies including Intel, AMD, IBM and Sun (Oracle) is presented. In addition to the overview, the main advantage of using multi-core will demonstrated by the experimental results. The focus of the experiment is to study speed-up in the execution of the ‘program’ as the number of the processors (core) increases. For this experiment, open source parallel program to count the primes numbers is considered and simulation are performed on 3 nodes Raspberry cluster . Obtained results show that execution time of the parallel program decreases as number of core increases.
    [Show full text]
  • Cielo Computational Environment Usage Model
    Cielo Computational Environment Usage Model With Mappings to ACE Requirements for the General Availability User Environment Capabilities Release Version 1.1 Version 1.0: Bob Tomlinson, John Cerutti, Robert A. Ballance (Eds.) Version 1.1: Manuel Vigil, Jeffrey Johnson, Karen Haskell, Robert A. Ballance (Eds.) Prepared by the Alliance for Computing at Extreme Scale (ACES), a partnership of Los Alamos National Laboratory and Sandia National Laboratories. Approved for public release, unlimited dissemination LA-UR-12-24015 July 2012 Los Alamos National Laboratory Sandia National Laboratories Disclaimer Unless otherwise indicated, this information has been authored by an employee or employees of the Los Alamos National Security, LLC. (LANS), operator of the Los Alamos National Laboratory under Contract No. DE-AC52-06NA25396 with the U.S. Department of Energy. The U.S. Government has rights to use, reproduce, and distribute this information. The public may copy and use this information without charge, provided that this Notice and any statement of authorship are reproduced on all copies. Neither the Government nor LANS makes any warranty, express or implied, or assumes any liability or responsibility for the use of this information. Bob Tomlinson – Los Alamos National Laboratory John H. Cerutti – Los Alamos National Laboratory Robert A. Ballance – Sandia National Laboratories Karen H. Haskell – Sandia National Laboratories (Editors) Cray, LibSci, and PathScale are federally registered trademarks. Cray Apprentice2, Cray Apprentice2 Desktop, Cray C++ Compiling System, Cray Fortran Compiler, Cray Linux Environment, Cray SHMEM, Cray XE, Cray XE6, Cray XT, Cray XTm, Cray XT3, Cray XT4, Cray XT5, Cray XT5h, Cray XT5m, Cray XT6, Cray XT6m, CrayDoc, CrayPort, CRInform, Gemini, Libsci and UNICOS/lc are trademarks of Cray Inc.
    [Show full text]
  • ECP Software Technology Capability Assessment Report
    ECP-RPT-ST-0001-2018 ECP Software Technology Capability Assessment Report Michael A. Heroux, Director ECP ST Jonathan Carter, Deputy Director ECP ST Rajeev Thakur, Programming Models & Runtimes Lead Jeffrey Vetter, Development Tools Lead Lois Curfman McInnes, Mathematical Libraries Lead James Ahrens, Data & Visualization Lead J. Robert Neely, Software Ecosystem & Delivery Lead July 1, 2018 DOCUMENT AVAILABILITY Reports produced after January 1, 1996, are generally available free via US Department of Energy (DOE) SciTech Connect. Website http://www.osti.gov/scitech/ Reports produced before January 1, 1996, may be purchased by members of the public from the following source: National Technical Information Service 5285 Port Royal Road Springfield, VA 22161 Telephone 703-605-6000 (1-800-553-6847) TDD 703-487-4639 Fax 703-605-6900 E-mail [email protected] Website http://www.ntis.gov/help/ordermethods.aspx Reports are available to DOE employees, DOE contractors, Energy Technology Data Exchange representatives, and International Nuclear Information System representatives from the following source: Office of Scientific and Technical Information PO Box 62 Oak Ridge, TN 37831 Telephone 865-576-8401 Fax 865-576-5728 E-mail [email protected] Website http://www.osti.gov/contact.html This report was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights.
    [Show full text]
  • Stepping Towards a Noiseless Linux Environment
    ROSS 2012 | June 29 2012 | Venice, Italy STEPPING TOWARDS A NOISELESS LINUX ENVIRONMENT Hakan Akkan*, Michael Lang¶, Lorie Liebrock* Presented by: Abhishek Kulkarni¶ * New Mexico Tech ¶ Ultrascale Systems Research Center New Mexico Consortium Los Alamos National Laboratory 29 June 2012 Stepping Towards A Noiseless Linux Environment 2 Motivation • HPC applications are unnecessarily interrupted by the OS far too often • OS noise (or jitter) includes interruptions that increase an application’s time to solution • Asymmetric CPU roles (OS cores vs Application cores) • Spatio-temporal partitioning of resources (Tessellation) • LWK and HPC Oses improve performance at scale 29 June 2012 Stepping Towards A Noiseless Linux Environment 3 OS noise exacerbates at scale • OS noise can cause a significant slowdown of the app • Delays the superstep since synchronization must wait for the slowest process: max(wi) Image: The Case of the Missing Supercomputer Performance, Petrini et. Al, 2003 29 June 2012 Stepping Towards A Noiseless Linux Environment 4 Noise co-scheduling • Co-scheduling the noise across the machine so all processes pay the price at the same time Image: The Case of the Missing Supercomputer Performance, Petrini et. Al, 2003 29 June 2012 Stepping Towards A Noiseless Linux Environment 5 Noise Resonance • Low frequency, Long duration noise • System services, daemons • Can be moved to separate cores • High frequency, Short duration noise • OS clock ticks • Not as easy to synchronize - usually much more frequent and shorter than the computation granularity of the application • Previous research • Tsafrir, Brightwell, Ferreira, Beckman, Hoefler • Indirect overhead is generally not acknowledged • Cache and TLB pollution • Other scalability issues: locking during ticks 29 June 2012 Stepping Towards A Noiseless Linux Environment 6 Some applications are memory and network bandwidth limited! *Sancho, et.
    [Show full text]
  • Nascent Exascale Supercomputers Offer Promise, Present Challenges CORE CONCEPTS Adam Mann, Science Writer
    CORE CONCEPTS Nascent exascale supercomputers offer promise, present challenges CORE CONCEPTS Adam Mann, Science Writer Sometime next year, managers at the US Department Laboratory in NM. “We have to change our computing of Energy’s (DOE) Argonne National Laboratory in paradigms, how we write our programs, and how we Lemont, IL, will power up a calculating machine the arrange computation and data management.” size of 10 tennis courts and vault the country into That’s because supercomputers are complex a new age of computing. The $500-million main- beasts, consisting of cabinets containing hundreds of frame, called Aurora, could become the world’sfirst thousands of processors. For these processors to oper- “exascale” supercomputer, running an astounding ate as a single entity, a supercomputer needs to pass 1018, or 1 quintillion, operations per second. data back and forth between its various parts, running Aurora is expected to have more than twice the huge numbers of computations at the same time, all peak performance of the current supercomputer record while minimizing power consumption. Writing pro- holder, a machine named Fugaku at the RIKEN Center grams for such parallel computing is not easy, and the- for Computational Science in Kobe, Japan. Fugaku and orists will need to leverage new tools such as machine its calculation kin serve a vital function in modern learning and artificial intelligence to make scientific scientific advancement, performing simulations crucial breakthroughs. Given these challenges, researchers for discoveries in a wide range of fields. But the transition have been planning for exascale computing for more to exascale will not be easy.
    [Show full text]
  • Experimental and Analytical Study of Xeon Phi Reliability
    View metadata, citation and similar papers at core.ac.uk brought to you by CORE provided by Lume 5.8 Experimental and Analytical Study of Xeon Phi Reliability Daniel Oliveira Laércio Pilla Nathan DeBardeleben Institute of Informatics, UFRGS Department of Informatics and Los Alamos National Laboratory Porto Alegre, RS, Brazil Statistics, UFSC Los Alamos, NM, US Florianópolis, SC, Brazil Sean Blanchard Heather Quinn Israel Koren Los Alamos National Laboratory Los Alamos National Laboratory University of Massachusetts, UMass Los Alamos, NM, US Los Alamos, NM, US Amherst, MA, US Philippe Navaux Paolo Rech Institute of Informatics, UFRGS Institute of Informatics, UFRGS Porto Alegre, RS, Brazil Porto Alegre, RS, Brazil ABSTRACT 1 INTRODUCTION We present an in-depth analysis of transient faults effects on HPC Accelerators are extensively used to expedite calculations in large applications in Intel Xeon Phi processors based on radiation experi- HPC centers. Tianhe-2, Cori, Trinity, and Oakforest-PACS use Intel ments and high-level fault injection. Besides measuring the realistic Xeon Phi and many other top supercomputers use other forms of error rates of Xeon Phi, we quantify Silent Data Corruption (SDCs) accelerators [17]. The main reasons to use accelerators are their by correlating the distribution of corrupted elements in the out- high computational capacity, low cost, reduced per-task energy put to the application’s characteristics. We evaluate the benefits consumption, and flexible development platforms. Unfortunately, of imprecise computing for reducing the programs’ error rate. For accelerators are also extremely likely to experience transient errors example, for HotSpot a 0.5% tolerance in the output value reduces as they are built with cutting-edge technology, have very high the error rate by 85%.
    [Show full text]