Linpack Linpack 1970 1980 Fortran Jack Dongarra Linpack

  • Slides: 29
Download presentation

Linpack Linpack, библиотека для решения систем линейных алгебраических уравнений 1970 -1980 е Fortran, Jack

Linpack Linpack, библиотека для решения систем линейных алгебраических уравнений 1970 -1980 е Fortran, Jack Dongarra Linpack Benchmark Ax = b FLOPS - FLoating point Operations Per Secon 2/3·N 3+ 2·N 2 операций Rpeak и Rmax

Ноябрь 2010 № Rmax Rpeak (Tflops) Название Компьютер Число процессорных ядер Страна 1 2566.

Ноябрь 2010 № Rmax Rpeak (Tflops) Название Компьютер Число процессорных ядер Страна 1 2566. 00 4701. 00 Tianhe-1 A NUDT TH MPP 186368 (X 5670 2. 93 Ghz 6 C, NVIDIA GPU, FT-1000 8 C) Китай 2 1759. 00 2331. 00 Jaguar Cray XT 5 224162 (Opteron) США 3 1271. 00 2984. 30 Nebulae Dawning TC 3600 Blade System 120640 (Xeon/NVidia Tesla) Китай 4 1192. 00 2287. 63 TSUBAME 2. 0 HP Pro. Liant SL 390 s G 7 73278 (Xeon 6 C X 5670, Nvidia GPU) Япония 5 1054. 00 1288. 63 Hopper Cray XE 6 12 -core 2. 1 GHz 153408 США 6 1050. 00 1254. 55 Tera-100 Bull bullx super-node S 6010/S 6030 138368 Франция 7 1042. 00 1375. 78 Roadrunner Blade. Center QS 22/LS 21 122400 (Cell/Opteron) США 8 831. 70 1028. 85 Kraken Cray XT 5 98928 (Opteron) США 9 825. 50 1002. 70 JUGENE Blue Gene/P Solution 294912 (POWER) Германия 10 816. 60 Cielo Cray XE 6 8 -core 2. 4 GHz 107152 США

Китай (5 лучших) Rank Site System Cores Rmax Rpeak 1 National Supercomputing Center in

Китай (5 лучших) Rank Site System Cores Rmax Rpeak 1 National Supercomputing Center in Tianjin China NUDT TH MPP, X 5670 2. 93 Ghz 6 C, NVIDIA GPU, FT-1000 8 C NUDT 186368 2566 4701 3 National Supercomputing Centre in Shenzhen (NSCS) China Dawning TC 3600 Blade, Intel X 5650, NVidia Tesla C 2050 GPU Dawning 120640 1271 2984. 3 28 Institute of Process Engineering, Chinese Academy of Sciences China Mole-8. 5 Cluster Xeon L 5520 2. 26 Ghz, n. Vidia Tesla, Infiniband IPE, Nvidia, Tyan 33120 207. 3 1138. 44 35 Shanghai Supercomputer Center China Dawning 5000 A, QC Opteron 1. 9 Ghz, Infiniband, Windows HPC 2008 Dawning 30720 180. 6 233. 47 68 Computer Network Information Center, Chinese Academy of Science China Deep. Comp 7000, HS 21/x 3950 Cluster, 12216 Xeon QC HT 3 GHz/2. 93 GHz, Infiniband Lenovo 102. 8 145. 97

Россия в топ 500 Ran k Site System Cores Rmax Rpeak 17 Moscow State

Россия в топ 500 Ran k Site System Cores Rmax Rpeak 17 Moscow State University - Research Computing Center Russia T-Platforms T-Blade 2, Xeon 5570 2. 93 GHz, Infiniband QDR T-Platforms 35360 350. 1 414. 42 62 Joint Supercomputer Center Russia Cluster Platform 3000 BL 460 c/BL 2 x 220, Xeon 54 xx 3 Ghz, Infiniband Hewlett-Packard 11680 107. 45 140. 16 71 Kurchatov Institute Moscow Russia Cluster Platform 3000 BL 2 x 220, E 54 xx 3. 0 Ghz, Infiniband Hewlett-Packard 10304 101. 21 123. 65 179 Moscow State University - Research Computing Center Russia T-Platforms T 60, Intel Quadcore 3 Ghz, Infiniband DDR SKIF/T-Platforms 5000 47. 17 60 202 Government Russia Blade. Center HS 22 Cluster, Xeon QC GT 2. 53 GHz, Infiniband IBM 4864 44. 82 49. 22 258 Government Russia x. Series x 3650 Cluster Xeon QC GT 2. 66 GHz, Infiniband IBM 4384 40. 39 46. 75 299 Banking Russia Blade. Center HS 22 Cluster, Xeon QC GT 2. 53 GHz, Gig. Ethernet IBM 6600 38. 1 66. 79 305 Classified Russia x. Series x 3650 M 2 Cluster, Xeon QC E 55 xx 2. 53 Ghz, Gig. E IBM 6496 37. 5 65. 74 329 IT Services Russia Blade. Center HS 22 Cluster, Xeon QC E 5 xx 2. 53 GHz, Gig. Ethernet IBM 6344 36. 62 64. 2 423 Government Russia x. Series x 3650 M 2 Cluster Xeon QC GT 2. 53 GHz, Gig. Ethernet IBM 5656 32. 65 57. 24 460 Government Russia Blade. Center HS 22 Cluster, Xeon QC E 5 xx 2. 53 GHz, Gig. Ethernet IBM 5496 31. 72 55. 62

Processor Generation

Processor Generation

Interconnect Family

Interconnect Family

Hopper (5 е место) Site: DOE/SC/LBNL/NERSC System Family: Cray XE System Model: Cray XE

Hopper (5 е место) Site: DOE/SC/LBNL/NERSC System Family: Cray XE System Model: Cray XE 6 Computer: Cray XE 6 12 -core 2. 1 GHz Vendor: Cray Inc. Application area: Research Installation: September 2010 Operating System: Linux Interconnect: Custom

TSUBAME 2. 0 (4 е место) Site GSIC Center, Tokyo Institute of Technology System

TSUBAME 2. 0 (4 е место) Site GSIC Center, Tokyo Institute of Technology System Family HP Cluster Platform 3000 SL System Model Cluster Platform SL 390 s G 7 Computer HP Pro. Liant SL 390 s G 7 Xeon 6 C X 5670, Nvidia GPU, Linux/Windows Vendor. NEC/HP Application area Research Installation Year 2010 Operating System Linux

Nebulae (3 е место) Site: National Supercomputing Centre in Shenzhen (NSCS) System Family: Dawning

Nebulae (3 е место) Site: National Supercomputing Centre in Shenzhen (NSCS) System Family: Dawning Cluster System Model: Dawning TC 3600 Blade System Computer: Dawning TC 3600 Blade, Intel X 5650, NVidia Tesla C 2050 GPU Vendor: Dawning Application area: Research Installation Year: 2010

Jaguar (2 е место) Site: DOE/SC/Oak Ridge National Laboratory System Family: Cray XT System

Jaguar (2 е место) Site: DOE/SC/Oak Ridge National Laboratory System Family: Cray XT System Model: Cray XT 5 -HE Computer: Cray XT 5 -HE Opteron 6 -core 2. 6 GHz Vendor: Cray Inc. Application area: Not Specified Installation Year: 2009 Operating System: Linux (CLE) Interconnect: Proprietary Processor: AMD x 86_64 Opteron Six Core 2600 MHz (10. 4 GFlops)

Tianhe-1 A [天河一号] (1 е место) Site: National Supercomputing Center in Tianjin System Model:

Tianhe-1 A [天河一号] (1 е место) Site: National Supercomputing Center in Tianjin System Model: NUDT YH MPP Computer: NUDT YH MPP, X 5670 2. 93 Ghz 6 C, NVIDIA GPU, FT-1000 8 C Main Memory: 229376 GB Installation Year: 2010 Operating System: Linux Memory: 229376 GB

Tianhe-1 4, 096 Intel Xeon E 5540 1, 024 Intel Xeon E 5450 5,

Tianhe-1 4, 096 Intel Xeon E 5540 1, 024 Intel Xeon E 5450 5, 120 AMD GPU, 2. 560 dual-GPU ATI Radeon HD 4870 X 2 Tianhe-1 A 14, 336 Xeon X 5670 (32 -nm, 2, 93 GHz, 6 cores, L 3 12 Mb, 1333 Mhz DDR 3) 7, 168 Nvidia Tesla M 2050 GPU (1. 15 GHz)

Tianhe 1 A System is composed of 112 computer cabinets 12 storage cabinets 6

Tianhe 1 A System is composed of 112 computer cabinets 12 storage cabinets 6 communications cabinets 8 I/O cabinets Each computer cabinet is composed of four frames Frame containing eight blades plus a 16 -port switching board. Each blade is composed of two computer nodes

Green 500 Rank MFLOPS/W Site* Computer* Total Power (k. W) 1 1684. 20 IBM

Green 500 Rank MFLOPS/W Site* Computer* Total Power (k. W) 1 1684. 20 IBM Thomas J. Watson Research Center NNSA/SC Blue Gene/Q Prototype 38. 80 5 773. 38 Forschungszentrum Juelich (FZJ) QPACE SFB TR Cluster, Power. XCell 8 i, 3. 2 GHz, 3 D-Torus 57. 54 11 635. 15 National Supercomputing Center in Tianjin NUDT YH Cluster, X 5670 2. 93 Ghz 6 C, NVIDIA GPU, FT-1000 8 C 4040. 00 14 492. 64 National Supercomputing Centre in Shenzhen (NSCS) Dawning TC 3600 Blade, Intel X 5650, NVidia Tesla C 2050 GPU 2580. 00 88 253. 07 DOE/SC/Oak Ridge National Laboratory Cray XT 5 -HE Opteron 6 -core 2. 6 GHz 6950. 60 101 230. 42 Moscow State University - Research Computing Center T-Platforms T-Blade 2, Xeon 5570 2. 93 GHz, Infiniband QDR 1519. 38

Green 500 Rank MFLOPS/W Site* Computer* Total Power (k. W) 1 1684. 20 IBM

Green 500 Rank MFLOPS/W Site* Computer* Total Power (k. W) 1 1684. 20 IBM Thomas J. Watson Research Center NNSA/SC Blue Gene/Q Prototype 38. 80 2 1448. 03 National Astronomical Observatory of Japan GRAPE-DR accelerator Cluster, Infiniband 24. 59 3 1031. 92 GSIC Center, Tokyo Institute of Technology TSUBAME 2. 0 Windows HPC Server Power. Efficiency Experimental 26. 00 4 933. 06 NCSA/UIUC Eco. G 36. 00 5 886. 07 CASPUR Jazz Cluster; DL 360 G 7, Tesla S 2070, HPC 2008 S_R 2 26. 00 6 828. 67 RIKEN Advanced Institute for Computational Science K computer, SPARC 64 VIIIfx 2. 0 GHz, Tofu interconnect 57. 96 7 773. 38 Forschungszentrum Juelich (FZJ) QPACE SFB TR Cluster, Power. XCell 8 i, 3. 2 GHz, 3 D-Torus 57. 54 7 773. 38 Universitaet Regensburg QPACE SFB TR Cluster, Power. XCell 8 i, 3. 2 GHz, 3 D-Torus 57. 54 7 773. 38 Universitaet Wuppertal QPACE SFB TR Cluster, Power. XCell 8 i, 3. 2 GHz, 3 D-Torus 57. 54 10 740. 78 Universitaet Frankfurt Supermicro Cluster, QC Opteron 2. 1 GHz, ATI Radeon GPU, Infiniband 385. 00

Graph 500 Rank Machine Owner Problem Size TEPS Implementation 1 DOE/SC/ANL Intrepid (IBM Blue.

Graph 500 Rank Machine Owner Problem Size TEPS Implementation 1 DOE/SC/ANL Intrepid (IBM Blue. Gene/P, 8192 of 40960 nodes / 32 k of 163840 cores) Argonne National Laboratory Scale 36 (Medium) 6. 6 GE/s Optimized 2 Franklin (Cray XT 4, 500 of 9544 nodes) NERSC Scale 32 (Small) 5. 22 GE/s Optimized 3 cougarxmt (128 node Cray XMT) Pacific Northwest National Laboratory Scale 29 (Mini) 1. 22 GE/s Optimized 4 graphstorm (128 node Cray XMT) Sandia National Laboratories Scale 29 (Mini) 1. 17 GE/s Optimized 5 Endeavor (256 node, 512 core Westmere X 5670 2. 93, IB network) Intel Corporation Scale 29 (Mini) 533 ME/s Reference 6 Erdos (64 node Cray XMT) Oak Ridge National Laboratory Scale 29 (Mini) 50. 5 ME/s Reference 7 Red Sky (Nehalem X 5570 @2. 93 GHz, IB Torus, 512 processors) Sandia National Laboratories Scale 28 (Toy++) 477. 5 ME/s Reference 8 Jaguar (Cray XT 5 -HE, 512 node subset) Oak Ridge National Laboratory Scale 27 (Toy+) 800 ME/s Reference 9 Endeavor (128 node, 256 core Westmere X 5670 2. 93, IB network) Intel Corporation Scale 26 (Toy) 615. 8 ME/s Reference