创新释放高性能计算潜力 - 华为 林俊

2020-02-27 990浏览

  • 1.创新释放高性能计算潜力 林俊：华为服务器领域首席架构师
  • 2.Market Trends 2 2
  • 3.Requirement for Compute Security Internet of Things Industry 4.0 Intelligent City Big Data 2020 Millions of MIPS Cloud Mobility Opportunity for Innovation Traditional Architecture 1989 20 MIPS 1972 0.004 MIPS 3 3 2014 124,000 MIPS
  • 4.Computing Innovation slows down Past Fast Growth High Power Single Core Now Multi Core Performance Increase Performance Limited is Slowing Down by mdahl’s Law限制 The Doubling of Transistors are Slowing Down 4 Low Utilization 4 Not Secure Uneven Subsystem Development
  • 5.The End of Moore’s Law 10 µm – 1971 6 µm – 1974 3 µm – 1977 1.5 µm – 1982 1 µm – 1985 800 nm – 1989 600 nm – 1994 350 nm – 1995 250 nm – 1997 180 nm – 1999 130 nm – 2001 90 nm – 2004 65 nm – 2006 45 nm – 2008 32 nm – 2010 22 nm – 2012 14 nm – 2014 10 nm – 2016 7 nm – 2018 5 nm – 2020 Process Covalent radius of Silicon Atom is 111 pm (0.111 nm) 5 5 Tock Tick Architecture Optimization
  • 6.Changes to CPU Power Consumption 6 6
  • 7.Increase usage of Accelerators Adoption accelerated since 2010. 7 7 Nvidia still dominates Accelerator-based system projected to dominate for next decade Total performance share plateaued in past year, mainly due to life cycle
  • 8.Heterogeneous Architecture Processors are moving toward specialization Heterogeneous CPUs can be more flexible, higher cost performance, and high power performance • First used by storage systems • Internet server begin small scale deployment from 2013～2014 • Enterprise server application still lag behind 3~5 years • Performance per Watt is becoming more important 8 8
  • 9.Huawei HPC Solution Technology 9 9
  • 10.World Class HPC Solutions TODAY 36 170+ $63B 14.2% 79,000 16 Countries 2015 Revenue % of Revenue in R&D R&D Engineers R&D Centers Joint Innovation Centers Huawei FusionServer Standalone Compute Node 1P-32P Modular HPC Systems 》 Petascale 》 Workload 》 Reduce 》 Design System Direct Liquid Cooling OceanStor NVMe SSD 10 HPC storage Optimization Ecosystem Partnership CloudEngine Big Data storage Network Fabric 10 Modular & Container Data center Complexity More Performance / $ for Growth HPC Private Cloud
  • 11.Simplify HPC Systems TODAY MAXIMIZE EFFICIENCY ACCELERATE WORKLOAD CONVERGED HPC & BIG DATA SDI SDS Big Data MORE COMPUTE LESS SPACE LOWER POWER CONSUMPTION 11 MAXIMIZE PERFORMANCE FOR INDIVIDUAL WORKLOAD  End-to-end energy efficient design  HVDC  Flexible, modular architecture  High Ambient Temperature ~40oC  Multiple innovative form factors  Direct Liquid coverage  Deep optimization with hardware acceleration  Tight integration with Huawei data center infrastructure  Super Fat nodes Cooling 11 ~ 84% MAXIMIZE HARDWARE ROI  Single HPC cluster and storage system for both traditional HPC MPI workload and Hadoop  Innovative big data analytic appliance with deep hardware and software optimization for maximum cost effectiveness
  • 12.HPC / IT Solutions for Tomorrow New Technology Enablement New Heterogeneous CPU Creative Converged Solutions Leverage Cross Disciplinary Assets Enabling HPC Cloud GPU/FPGA Accelerato r Next Gen CPU HBM/HMC DDR4 FusionInsight Big Data 〉〉〉〉 Big Data Acceleration GPU/FPGA Accelerato r HBM/HMC ManageOne Management software 〉〉〉〉 New Memory Hierarchy FusionSphere Cloud OS 〉〉〉〉 DDR4-SCM-DIMM DDR4 DRAM as SCM Cache SCM-SSD 12 SRAM Cache HB M GPU X86 CPU ME M FPGA NVMeSSD Deep Learning FusionStorage Software Defined Storage Pool 〉〉〉〉 HBA/RAID HDD 12
  • 13.HPC Cloud Framework Agile - Open architecture Technical Computing User Login Graphic Virtualized Distributed File System Simulation Resource Pool GPGPU Object Store - Rapid deployment Graphic Visualization - Efficient operation Head Node Job Scheduler DB Compute Node CPU Intense Memory Intense Storage Node HPC NAS HPC Block Elastic - Demand based - Maximize utilization Secure - Multi-tenents VPC isolation HPC Compute 13 HPC Distributed Storage 13 Low Latency Networking Security Isolation - On-premise to cloud end-to-end secured
  • 14.Converged HPC & Big Data 14 14
  • 15.Builds A Leading Computing Platform With NVIDIA 15 15
  • 16.HPC Solutions THANK YOU Copyright©2016 Huawei Technologies Co., Ltd. All Rights Reserved. The information in this document may contain predictive statements including, without limitation, statements regarding the future financial and operating results, future product portfolio, new technology, etc. There are a number of factors that could cause actual results and developments to differ materially from those expressed or implied in the predictive statements. Therefore, such information is provided for reference purpose only and constitutes neither an offer nor an acceptance. Huawei may change the information at any time without notice.
  • 17.HUAWEI HPC Momentum Manufacturing CAE/CFD Energy Production & Distribution 17 17 Chip Design & Manufacturing Oil & Gas Exploration Education/Research/Supercomputing Digital Media
  • 18.Industrial CAE Simulation General development process disadvantages Crash & safety Indoor acoustics NVH Computational fluid dynamics (CFD) Planning Design Static strength Structural mechanics Sample Verification Electromagnetics Vibration and noise Electromagnetic simulation … Design CFD System engineering Sample Verification Product  Long development period  High design cost  Weak process control Customers' main requirements  Short development cycle  Low cost  Intuitive analysis and controllable process Industrial HPC cluster highlights Processing before modeling Computing resolution Processing after analysis    Physical component 18 Size model Computing model Computing analysis 18 Result obtaining Verification analysis Application integration and optimization Operation cost reduction Large-scale cluster management
  • 19.Huawei CAE Simulation Solution Industrial simulation application scenarios Crash simulation test Bright Computing Application optimization centralized management PARATERAComputing:'>Computing: