Yunquan Zhang Jiachang Sun Guoxing Yuan Linbo Zhang
Yunquan Zhang Jiachang Sun Guoxing Yuan Linbo Zhang Lab. Of Parallel Software and Computational Science, ISCAS Speaker: Liang Yuan Lab. Of Parallel Software and Computational Science, ISCAS Place photo here State-of-the-Art Analysis and Perspectives of China HPC Development: A View from 2010 HPC TOP 100
China HPC TOP 100: Background • First list published in 2002 • Funded by National 863 Plan in 2004 and afterwards • Selected by Chinese Science and Technology Annual Reports in 2005、2006、2007 • Referred by many international reports on China HPC study • Collaboration with TOP 500 • Keynotes presentations at US Supercomputing Workshop in 2007 and 2010
2010 China HPC TOP 100 Authors Yunquan Zhang, Jiachang Sun, Guoxin Yuan, Linbo Zhang The Specialty Association of Mathematical & Scientific Software (SAMSS) Evaluation Center of High Performance Computer, National 863 Plan China HPC Technical Committee
Remarks • Data source from Mainland China only • “Q”:Tested by SAMSS • “T”: From TOP 500 • “C”: From IHV • “U”: From Users • “S”: Extrapolated from similar system on TOP 500 • User is responsible for the accuracy of the data they provided. We just did sanity check • The list is published in fall every year
2010 China HPC Top 10 排 名 厂 商 Manufact urer 安装地点 Installation Site 安装年份 Year �理器 核 Num of Proc Linpack (Gflops) 1 国防科大 NUDT 国家超� � 算天津中心 2010 202, 752 2, 507, 000. 00 4, 701, 000. 00 0. 533 2 曙光 Dawning 曙光天津� �基地 2010 120, 640 1, 271, 000. 00 2, 984, 300. 00 0. 426 3 中科院� 程所 IPE, CAS Mole-8. 5 Cluster/320 x 2 Intel QC Xeon E 5520 2. 26 Ghz + 320 x 6 Nvidia Tesla C 2050/QDR Infiniband 中国科学院� 程 程研究所 2010 33, 120 207, 300. 00 1, 138, 440. 00 0. 182 4 曙光 Dawning 魔方/曙光5000 A/1920 x 4 AMD QC Barcelona 1. 9 GHz/DDR Infiniband/WCCS+Linux 上海超� � 算中心 2008 30, 720 180, 600. 00 233, 472. 00 0. 774 5 �想 Lenovo 中国科学院超 ��算中心 2008 12, 160 106, 500. 00 145, 293. 00 0. 733 6 曙光 Dawning 成都超�� 算中心(二期) 2010 5, 720 76, 350. 38 141, 389. 60 0. 540 7 曙光 Dawning 中国科学院� 算技�研究所 2010 4, 160 55, 527. 55 102, 828. 80 0. 540 8 IBM x. Series x 3650 M 2 Cluster/Intel Xeon QC E 55 xx 2. 53 Ghz/Giga-E 程公司 2010 8, 960 51, 200. 00 90, 680. 00 0. 565 9 HP Cluster Platform 3000 BL 460 c G 6/Intel Xeon E 5540 2. 53 GHz/Giga-E 中国�信 2010 7, 848 41, 880. 00 79, 420. 00 0. 527 10 IBM Blade. Center HS 22 Cluster/Intel Xeon QC GT 2. 53 GHz/Giga-E 网�公司 2009 7, 168 41, 270. 00 72, 540. 00 0. 569 型号 Computer 天河一号/Tianhe 1 A/7168 x 2 Intel Hexa Core Xeon X 5670 2. 93 GHz + 7168 Nvidia Tesla M 2050@1. 15 GHz+2048 Hex Core FT 1000@1 GHz/私有高速网� 80 Gbps 曙光星云/Dawning TC 3600 Blade/Intel Hexa Core X 5650 + Nvidia Tesla C 2050 GPU/QDR Infiniband 深� 7000/1240 x 2 Intel Xeon QC E 5450 3. 0 GHz/140 x 4 Intel Xeon QC X 7350 2. 93 GHz Infiniband 4 x. DDR 曙光星云/Dawning TC 3600 Blade/220 x(2 Intel Hexa Core X 5650 + 1 NVidia Tesla C 2050)/QDR Infiniband 生物�用机/Dawning TC 3600 Blade/Intel Hexa Core X 5650 + NVidia Tesla C 2050 GPU/QDR Infiniband Peak (Gflops) 效率 Efficiency
China HPC TOP 100 Authors with Tianhe 1 A
International Collaboration Prof. Jack Dongarra and Thomas Sterling
China HPC TOP 100 Performance Analysis • Tianhe 1 A from National University of Defense Technology takes #1 again with Linpack performance of 2. 5 Pflops • The Linpack performance of all systems is above 9. 6 TFlops • Peak performance all exceeds 11 TFlops Total Performance Ratio 7 6 5 4 • The first 3 systems are CPU/GPU hybrid • 98 out of 100 are clusters 3 2 1 0 2008 2009 2010
Cluster Share in China HPC TOP 100 0 20 1 9 20 0 8 20 0 7 20 0 6 20 0 5 20 0 4 20 0 3 20 0 2 100 90 80 70 60 50 40 30 20 10 0 20 0 Count Cluster Share
Manufacturer Analysis Domestic Manufacturer Systems Share Rmax [TF/s] Rpeak [TF/s] Dawning 34 34% 2028. 19 4218. 89 61. 07% 233436 Inspur 5 5% 92. 11 115. 38 78. 30% 10360 Lenovo 3 3% 126. 69 182. 27 50. 83% 16128 Sunway 3 3% 50. 74 64. 49 80. 23% 6096 Power. Leader 2 2% 40. 38 51. 20 79. 00% 4320 NUDT 1 1% 2507. 00 4701. 00 53. 30% 202752 IPE 1 1% 207. 30 1138. 44 18. 20% 33120 49 49% 5052. 41 10471. 67 60. 13% 506212 IBM 28 28% 753. 01 1328. 21 58. 13% 133000 HP 19 19% 367. 46 629. 12 60. 93% 65508 Dell 3 3% 47. 83 74. 60 72. 43% 6880 SUN 1 1% 10. 46 13. 58 66. 00% 1200 51 51% 1178. 76 2045. 51 64. 37% 206588 100% 6231. 17 12517. 59 62. 00% 712800 Domestic Total Import Total Average Efficiency Num of Core
Manufacturer Shares By Number of Systems HP, 19 Inspur, 5 DELL, 3 Lenovo, 3 Sunway, 3 Power. Leader, 2 NUDT, 1 SUN, 1 IPE, 1 IBM, 28 Dawning, 34 2010 China HPC TOP 100 http: //www. samss. org. cn
Manufacturer Share Trend Domestic Import 100 90 80 70 60 50 40 30 20 10 0 2002 2003 2004 2005 2006 2007 2008 2009 2010 HP IBM SGI SUN DELL 曙光Dawning �想Lenovo 神威Sunway 浪潮Inspur 清�大学Tsinghua Univ. 宝德Power. Leader �壳星盈Galactic 上海大学Shanghai Univ. 自行�装Self Assembled �云Huayun �算所ICT 聚星Juxin 北京�算中心Beijing Computer Center 其它Others
Manufacturer Share by Performance Dawning, 32. 55% IBM, 12. 08% HP, 5. 90% IPE, 3. 33% NUDT, 40. 23% 2010 China HPC TOP 100 http: //www. samss. org. cn Lenovo, 2. 03% Inspur, 1. 48% Sunway, 0. 81% DELL, 0. 77% Power. Leader, SUN, 0. 17% 0. 65%
Application Areas �用�域 Area 能源 Energy � Industry 科学�算 Research 游� Gaming 政府部� Government �信 Telecomm 教育 Education 气象 Weather 生物信息 Bio 互�网 Internet 后勤服� Logistics 地震 Earthquake ���算 Visualization �力 Power �漫渲染 DDC 物�网 Internet of Things 金融保� Finance �� Total 数量 份� Share # systems 17 15 12 9 9 7 7 5 4 4 2 2 1 100 17% 15% 12% 9% 9% 7% 7% 5% 4% 4% 2% 2% 1% 100% Linpack[GF/s] 峰� Peak [GF/s] 265508. 07 467189. 50 4299853. 48 8516574. 64 476779. 40 1491403. 64 291100. 00 517130. 00 138162. 97 266433. 60 187450. 40 348690. 34 129689. 42 167107. 76 85589. 00 115121. 52 100894. 55 178611. 80 88469. 25 163946. 00 43939. 10 81960. 96 37372. 00 50066. 08 31507. 37 58988. 16 21726. 15 38752. 00 12115. 26 22131. 20 11095. 04 20377. 60 9830. 25 13107. 00 6231171. 71 12517591. 80 平均效率 Efficiency 59. 07% 70. 76% 73. 83% 55. 76% 52. 07% 53. 84% 77. 94% 74. 62% 63. 03% 53. 40% 53. 95% 76. 15% 53. 40% 56. 15% 54. 70% 54. 40% 75. 00% 62. 00% �理器数 # of Proc 46100 401324 64376 51136 29096 37360 13624 12192 10864 16600 8368 4608 6608 4240 2080 2176 2048 712800
Application Area System Shares
Application Area Trend 100 80 60 40 20 Telcomm Gaming Graph 2010 2009 Industry Finacial Earthquake 2008 2007 Edu Bioinfo Transport 2006 Energy Gov Database 2005 2004 Science Computing Climate Tax 2003 2002 0
Application Area Performance Shares
Application Areas Analysis • Number of application areas exceeds previous years • Number of systems: Top areas are energy, industry, and research • Total system performance: Top areas are industry, research, and gaming • Main users: Energy, industry, research, gaming, and government • New users: Internet of things, internet, and power
Multicore Processor Shares 2 Core, 2%12 Core, 3% 6 Core, 14% 4 Core, 81% 2010 China HPC TOP 100 http: //www. samss. org. cn
Processor Manufacturer Shares IBM, 1% AMD, 19% Intel, 80% 2010 China HPC TOP 100 http: //www. samss. org. cn
Interconnect Shares Hyper. Plex, Federation, NUDT 10 GE, 1% 1% Proprietary, 1% 1% Infiniband, 37% Giga-E, 59%
Performance Trend
Trend & Outlook (1) • 1993 -2010 China HPC performance increase • 1993 -1996 Steady growth • 1996 -1999 A large leap • 1999 -2001 Steady growth • 2001 -2005 Another period of rapid increase • 2005 -2007 Steady growth again • After 2008, active development in the next 2 -3 years
Trend & Outlook (2) Previous Predictions V. S. Real • 2007 -2008: System with peak performance of 100 TFlops (Reality: Oct 2008) • 2008 -2009: Total Linpack performance exceeds Pflops (Reality: Oct 2008) • 2010 -2011: System with peak performance of 1 PFlops (Reality: Oct 2009)
Trend & Outlook (3) Future Predictions • 2011 -2012: System with peak performance of 10 PFlops • 2012 -2013: Total Linpack performance reaches 10 PFlops • 2013 -2014: System with peak performance of 100 PFlops • 2014 -2015: Total Linpack performance reaches 100 PFlops
Thank You • Contact: Yunquan Zhang • Emails: zyq@mail. rdcps. ac. cn
- Slides: 26