SandyBridge v.s. Nehalem : Graph500

昨日の続きで、今回は Graph500 ベンチマークによる比較。SDPA では計算サーバ１と２では約２倍の性能差があったが、Graph500 では僅差となっている。

○計算サーバ１ (1 CPU x 4 コア = 4 コア)
CPU : Intel Core i7 2600 (3.40GHz / 8MB L3)
Memory : 16GB (4 x 4GB)
OS : Fedora 17 for x86_64

----------------------------------------------------------------------
Parallel Breadth-First Search for Graph500 Benchmark version 3.58
----------------------------------------------------------------------
CPU name is Intel(R) Core(TM) i7-2600 CPU @ 3.40GHz
freq / RAM is 3392.287 MHz / 15.59 GB
#cpu, #nodes, #cores is 8 1 8
COMPILER is GCC (GNU C Compiler) version 4.7.2
----------------------------------------------------------------------
scale, edgefactor is 23 16
energy_loop is disable
#threads, #NUMAs is 4 1
mpol_bind is ON(mmap with mbind(MPOL_BIND))
mem_interleave is OFF
switching parameter is 4 1
queue buffer size is 16384
----------------------------------------------------------------------
SCALE: 23
nvtx: 8388608
edgefactor: 16
terasize: 2.14748364799999996e-03
A: 5.69999999999999951e-01
B: 1.90000000000000002e-01
C: 1.90000000000000002e-01
D: 5.00000000000000444e-02
generation_time: 2.06679120063781738e+01
construction_time: 3.17045090198516846e+01
nbfs: 64
min_time: 6.96420669555664062e-02
firstquartile_time: 7.47668147087097168e-02
median_time: 8.54336023330688477e-02
thirdquartile_time: 1.00139021873474121e-01
max_time: 1.51937246322631836e-01
mean_time: 9.03971269726753235e-02
stddev_time: 2.07324407475770744e-02
min_nedge: 1.34216250000000000e+08
firstquartile_nedge: 1.34216250000000000e+08
median_nedge: 1.34216250000000000e+08
thirdquartile_nedge: 1.34216250000000000e+08
max_nedge: 1.34216250000000000e+08
mean_nedge: 1.34216250000000000e+08
stddev_nedge: 0.00000000000000000e+00
min_TEPS: 8.83366345306784749e+08
firstquartile_TEPS: 1.35747398808613539e+09
median_TEPS: 1.58310517464317441e+09
thirdquartile_TEPS: 1.80009671918441296e+09
max_TEPS: 1.92722955919205761e+09
harmonic_mean_TEPS: 1.48474021791168261e+09
harmonic_stddev_TEPS: 4.29018483066712841e+07


○計算サーバ２ (1 CPU x 4 コア = 4 コア)
CPU : Intel Core i7 860 (2.80GHz / 8MB L3)
Memory : 8GB (2 x 4GB)
OS : Fedora 17 for x86_64

----------------------------------------------------------------------
Parallel Breadth-First Search for Graph500 Benchmark version 3.58
----------------------------------------------------------------------
CPU name is Intel(R) Core(TM) i7 CPU 860 @ 2.80GHz
freq / RAM is 2806.964 MHz / 7.79 GB
#cpu, #nodes, #cores is 4 1 4
COMPILER is GCC (GNU C Compiler) version 4.7.2
----------------------------------------------------------------------
scale, edgefactor is 23 16
energy_loop is disable
#threads, #NUMAs is 4 1
mpol_bind is ON(mmap with mbind(MPOL_BIND))
mem_interleave is OFF
switching parameter is 4 1
queue buffer size is 16384
----------------------------------------------------------------------
SCALE: 23
nvtx: 8388608
edgefactor: 16
terasize: 2.14748364799999996e-03
A: 5.69999999999999951e-01
B: 1.90000000000000002e-01
C: 1.90000000000000002e-01
D: 5.00000000000000444e-02
generation_time: 3.14099318981170654e+01
construction_time: 4.24007842540740967e+01
nbfs: 64
min_time: 8.00127983093261719e-02
firstquartile_time: 8.69174003601074219e-02
median_time: 1.01752161979675293e-01
thirdquartile_time: 1.16612851619720459e-01
max_time: 1.82719945907592773e-01
mean_time: 1.06371581554412842e-01
stddev_time: 2.62080855166371765e-02
min_nedge: 1.34216250000000000e+08
firstquartile_nedge: 1.34216250000000000e+08
median_nedge: 1.34216250000000000e+08
thirdquartile_nedge: 1.34216250000000000e+08
max_nedge: 1.34216250000000000e+08
mean_nedge: 1.34216250000000000e+08
stddev_nedge: 0.00000000000000000e+00
min_TEPS: 7.34546244162514091e+08
firstquartile_TEPS: 1.16410484620945454e+09
median_TEPS: 1.33785618677348900e+09
thirdquartile_TEPS: 1.56306818916057491e+09
max_TEPS: 1.67743477088659644e+09
harmonic_mean_TEPS: 1.26176792747359514e+09
harmonic_stddev_TEPS: 3.91668721164102703e+07