GPU クラスタでの Graph500 性能 : GPU 使用

以下の 16GPU を用いて Graph500 の性能測定を行いました。

◯インテルコンパイラ 18.0.1
============= Result ==============
SCALE: 29
edgefactor: 16
NBFS: 16
graph_generation: 132.257772982
num_mpi_processes: 16
construction_time: 109.879436388
redistribution_time: 9.42433531606
min_time: 1.01756
firstquartile_time: 1.11051
median_time: 1.18029
thirdquartile_time: 1.24541
max_time: 1.607
mean_time: 1.19345
stddev_time: 0.141292
min_nedge: 8589858508
firstquartile_nedge: 8589858508
median_nedge: 8589858508
thirdquartile_nedge: 8589858508
max_nedge: 8589858508
mean_nedge: 8589858508
stddev_nedge: 0
min_TEPS: 5.34527e+09
firstquartile_TEPS: 6.89722e+09
median_TEPS: 7.27774e+09
thirdquartile_TEPS: 7.73507e+09
max_TEPS: 8.44163e+09
harmonic_mean_TEPS: 7.19752e+09
harmonic_stddev_TEPS: 2.20015e+08
min_validate: 1.8999e-07
firstquartile_validate: 1.96975e-07
median_validate: 2.3155e-07
thirdquartile_validate: 2.70084e-07
max_validate: 22.5079
mean_validate: 1.40674
stddev_validate: 5.62696

◯ gcc 4.8.5
============= Result ==============
SCALE: 29
edgefactor: 16
NBFS: 16
graph_generation: 75.078974192
num_mpi_processes: 16
construction_time: 106.847716701
redistribution_time: 9.1193040451
min_time: 1.04306
firstquartile_time: 1.13172
median_time: 1.1952
thirdquartile_time: 1.25732
max_time: 1.78853
mean_time: 1.22756
stddev_time: 0.175735
min_nedge: 8589858508
firstquartile_nedge: 8589858508
median_nedge: 8589858508
thirdquartile_nedge: 8589858508
max_nedge: 8589858508
mean_nedge: 8589858508
stddev_nedge: 0
min_TEPS: 4.80276e+09
firstquartile_TEPS: 6.83188e+09
median_TEPS: 7.18698e+09
thirdquartile_TEPS: 7.59008e+09
max_TEPS: 8.23527e+09
harmonic_mean_TEPS: 6.99751e+09
harmonic_stddev_TEPS: 2.58651e+08
min_validate: 9.40636e-08
firstquartile_validate: 1.20956e-07
median_validate: 1.38418e-07
thirdquartile_validate: 1.7439e-07
max_validate: 22.478
mean_validate: 1.40487
stddev_validate: 5.6195


◯追加分
ノード数４
各ノード
CPU : Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz x 2 : 10cores / 1CPU
GPU : NVIDIA K40m x 2
メモリ：256GB
ネットワーク：GbE x 2 + Mellanox Infiniband FDR x 2
HDD : HP 1000GB 7.2krpm SC 2.5型 6G SATA ハードディスクドライブ
OS : CentOS 7.4


◯現在の GPU クラスタ
ノード数４
各ノード
CPU : Intel(R) Xeon(R) CPU E5-2640 v3 @ 2.60GHz x 2 : 8cores / 1CPU
GPU : NVIDIA K40m x 2
メモリ：256GB
ネットワーク：GbE x 2 + Mellanox Infiniband FDR x 2
HDD : HP 500GB 7.2krpm SC 2.5型 6G SATA ハードディスクドライブ
OS : CentOS 7.4