Las redes
www.top500.org
NOVEMBER 2011
NAME/MANUFACTURER/COMPUTER SITE COUNTRY CORES
Rmax
Pflop/s
1 2 3 4 5
K computer SPARC64 VIIIfx 2.0GHz, Tofu interconnect Tianhe-1A 6-core IntelX5670 2.93 GHz + Nvidia M2050 GPU w/custom interconnect Jaguar Cray XT-5 6-core AMD 2.6 GHz w/custom interconnect Nebulae Dawning TC3600 Blade Intel X5650 2.67 GHz, NVidia Tesla C2050 GPU w/ IbandTsubame 2.0 HP Proliant SL390s G7 nodes (Xeon X5670 2.93GHz) , NVIDIA Tesla M2050 GPU w/Iband
RIKEN NUDT/NSCC/Tianjin DOE/OS/ORNL NSCS TiTech
Japan China USA China Japan
705,024 186,368 224,162120,640 73,278
10.5 2.57 1.76 1.27 1.19
PERFORMANCE DEVELOPMENT
1 Eflop/s 100 Pflop/s 10 Pflop/s 1 Pflop/s 100 Tflop/s 10 Tflop/s 1 Tflop/s 100 Gflop/s 10 Gflop/s 1 Gflop/s
Gflop/s
PROJECTED
74.2Pflop/s
10.5
SUM
Pflop/s
50.9
Tflop/s
1.17
Tflop/s
N=1
59.7
Gflop/s
N=500
0.4
’93
’94
’95
’96
’97
’98
’99
’00
’01
’02
’03
’04
’05
’06’07
’08
’09
’10
’11
’12
’13
’14
’15
’16
’17
’18
ARCHITECTURES
100% 100%
CHIP TECHNOLOGY
ALPHA
SIMD
80%
80%
CONSTELLATIONS MPP CLUSTERS
IBM
60%60%
HP MIPS INTEL
40%
40%
SMP
20% 20%
SPARC PROPRIETARY AMD
'98 '99 '00 '01 '02 '03 '04 '05 '06 '07 '08 '09 '10 ’11
SINGLE PROC.
‘93 ‘94 '95 '96 '97 '98 '99 '00 '01 '02 '03 '04'05 '06 '07 '08 '09 '10 ’11 ‘93 ‘94
'95
'96
'97
INSTALLATION TYPE
100%
VENDOR RESEARCH
80%
A Portable Implementation of the High Performance Linpack Benchmark for DistributedMemory Computers
INDUSTRY
60%
Algorithm: recursive panel factorizations, multiple lookahead depths, bandwidth reducing swapping Easy to install, only needs MPI + BLAS or VSIPL
40%
20%CLASSIFIED
GOVERNMENT
Highly scalable and efficient from the smallest cluster to the largest supercomputers in the world
FIND OUT MORE AT http://icl.eecs.utk.edu/hpl/
ACADEMIC
‘93 ‘94 '95...
Regístrate para leer el documento completo.