| JAC Benchmark (PME, 23.5K atm) |
Factor IX Benchmark (PME, 91K atm) |
| Cellulose Benchmark (PME, 408K atm) |
GB_MB Benchmark (GB, 2.5K atm) |
GB Nucleosome Benchmark (GB, 25K atm) |
Notes:
==============================================================================================
This is the protein DHFR, solvated with TIP3 water, in a periodic box.
There are 23,558 total atoms, and PME used with a direct space cutoff of 9 Ang.
This is the benchmark in benchmarks/jac subdirectory of the Amber 9 distribution.
---------------------------------------------------------------------------------------------
name date CPU OS compiler npcu ps per day speedup
---------------------------------------------------------------------------------------------
Datastar 04/06 1.5GHz IBM P655 AIX XLF90 1 197.30
(SDSC) 2 392.68 1.99
(IBM Federation Switch) 4 744.69 3.77
8 1397.29 7.08
16 2616.44 13.3
32 4851.21 24.6
64 8368.85 42.4
128 12087.30 61.3
192 12910.94 65.4
256 13688.21 69.4
384 10775.75
Datastar 04/06 1.7GHz IBM P655+ AIX XLF90 1 227.47
(SDSC) 2 454.22 2.00
(IBM Federation Switch) 4 862.76 3.79
8 1600.18 7.03
16 3008.57 13.2
32 5608.93 24.7
64 9602.13 42.2
128 14154.65 62.2
192 14285.71 62.8
256 15428.57 67.8
384 12206.84
Teragrid 04/06 1.5GHz Itanium 2 RH AS4 ifort 1 395.37
(SDSC) (9.0.033) 2 667.00 1.69
(Myrinet) (MKL8.0) 4 1200.97 3.04
8 2102.80 5.32
16 3008.57 7.61
32 3882.10 9.82
64 6115.52 15.5
128 7948.48 20.1
192 11131.15 28.2
256 10611.64
caffeine 04/06 3.2Ghz Pentium D RH AS4 ifort 1 231.79
(SDSC) (Dual Core) (9.0.033) 2 438.11
(MKL8.0, MPICH2)
coffee 04/06 2.8Ghz Pentium 4 RH AS4 ifort 1 192.82
(SDSC) Single Core HT (9.0.033) 2* 176.55*
(MKL8.0, MPICH2)
--------------------------------------------------------------------------------
* = Hyperthreading (2 = 1 Real + 1HT)
================================================================================
"fix": This is a factor_ix benchmark, from Bob Duke, also a protein
solvated with TIP3 water, in a periodic box. There are 90,906 total atoms,
and PME used with a direct space cutoff of 8 Ang. This test is in
amber9/benchmarks/factor_ix. This uses dt=0.0015, so the conversion is
ps-per-day = 129.6/(time-per-step)
--------------------------------------------------------------------------------
name date CPU OS compiler npcu ps per day
--------------------------------------------------------------------------------
Datastar 04/06 1.5GHz IBM PWR4 AIX XLF90 1 96.67
(SDSC) 2 186.15
(IBM Federation Switch) 4 354.11
8 634.73
16 1218.56
32 2292.59
64 4120.17
128 6822.85
192 8022.28
256 9243.94
384 9123.55
Datastar 04/06 1.7GHz IBM PWR4 AIX XLF90 1 110.17
(SDSC) 2 212.71
(IBM Federation Switch) 4 398.11
8 719.46
16 1380.19
32 2596.15
64 4677.01
128 7772.11
192 9784.82
256 10464.27
384 10160.72
Teragrid 04/06 1.5GHz Itanium 2 RH AS4 ifort 1 176.11
(SDSC) (9.0.033) 2 275.74
(Myrinet) (MKL8.0) 4 518.00
8 935.50
16 1606.55
32 2652.21
64 3887.81
128 5737.05
192 6476.14
Bluegene/L 04/06 0.7GHz PowerPC Custom OS xlf90 1 28.73 (CO mode)
(SDSC) (440d) 2 57.32 1.99
(custom torus and tree) (VN mode) 4 110.17 3.83
8 210.81 7.34
16 407.18 14.17
32 739.73 25.74
64 1353.10 47.09
128 2123.55 73.91
192 2495.19 86.84
256 2966.01 103.23
384 3139.16 109.25
512 2997.92 104.34
caffeine 04/06 3.2Ghz Pentium D RH AS4 ifort 1 111.96
(SDSC) (Dual Core) (9.0.033) 2 207.63
(MKL8.0, MPICH2)
coffee 04/06 2.8Ghz Pentium 4 RH AS4 ifort 1 97.16
(SDSC) Single Core HT (9.0.033) 2* 103.52*
(MKL8.0, MPICH2)
--------------------------------------------------------------------------------
* = Hyperthreading (2 = 1 Real + 1HT)
===========================================================================================
This is a benchmark of a cellulose fibre solvated in TIP3P water in a periodic
box. This is a large simulation with over 408,000 atoms, hence the scaling is
might be better than for the smaller benchmarks.
PME is used with a direct space cutoff of 8 Ang, Nrespa=2.
------------------------------------------------------------------------------------------
name date CPU OS compiler npcu ps per day
------------------------------------------------------------------------------------------
Datastar 04/06 1.5GHz IBM P655 AIX XLF90 1 14.44
(SDSC) 2 28.01 1.94
(IBM Federation Switch) 4 53.96 3.74
8 103.03 7.14
16 197.27 13.7
32 371.40 25.7
64 633.87 43.9
128 945.14 65.4
192 1201.50 83.2
256 1342.24 93.0
384 1521.53 105.4
512 1664.42 115.2
768 2622.16 181.6
1024 2558.48
Teragrid 04/06 1.5GHz Itanium 2 RH AS4 ifort 1 28.41
(SDSC) (9.0.033) 2 45.90
(Myrinet) (MKL8.0) 4 87.52
8 162.61
16 298.53
32 530.92
64 792.22
128 915.74
192 1153.31
256 1427.86
384 1545.76
Bluegene/L 04/06 0.7GHz PowerPC Custom OS xlf90 1 4.55 (CO mode)
(SDSC) (440d) 2 9.10 2.00(CO mode)
(custom torus and tree) (VN mode) 4 17.71 3.89(CO mode)
8 34.09 7.49(CO mode)
16 64.44 14.15
32 124.16 25.27
64 215.23 47.28
128 283.97 62.38
192 391.71 86.04
256 486.79 106.93
384 660.02 144.98
512 739.69 162.48
768 914.77 200.94
1024 891.59 195.85
------------------------------------------------------------------------------------------
=============================================================================================
"gb_mb" == Generalized Born myoglobin simulation. This protein has 2492
atoms, and is run with a 20 Ang. cutoff and a salt concentration of 0.2 M,
with nrespa=4 (long range forces computed every 4 steps.) This is the
test case in the benchmarks/gb_mb subdirectory of the Amber 9 distribution.
Note: PMEMD 9.0 supports GB simulations so these timings are for PMEMD.
---------------------------------------------------------------------------------------------
name date CPU OS compiler npcu ps per day speedup
---------------------------------------------------------------------------------------------
Datastar 04/06 1.5GHz IBM PWR4 AIX XLF90 1 220.14
(SDSC) 2 430.69 1.96
(IBM Federation Switch) 4 855.55 3.89
8 1676.76 7.62
16 3243.97 14.7
32 5992.51 27.2
64 9869.77 44.8
128 14257.43 64.7
Datastar 04/06 1.7GHz IBM PWR4 AIX XLF90 1 249.93
(SDSC) 2 488.04
(IBM Federation Switch) 4 955.14
8 1916.85
16 3696.42
32 6858.23
64 10701.02
128 15882.35
Teragrid 04/06 1.5GHz Itanium 2 RH AS4 ifort 1 191.51
(SDSC) (9.0.033) 2 358.97 1.87
(Myrinet) (MKL8.0) 4 789.85 4.12
8 1524.46 7.96
16 2839.12 14.8
32 4830.59 25.2
64 7651.44 40.0
128 10695.72 55.8
caffeine 04/06 3.2Ghz Pentium D RH AS4 ifort 1 266.03
(SDSC) (Dual Core) (9.0.033) 2 530.88
(MKL8.0, MPICH2)
coffee 04/06 2.8Ghz Pentium 4 RH AS4 ifort 1 239.89
(SDSC) Single Core HT (9.0.033) 2* 213.80*
(MKL8.0, MPICH2)
---------------------------------------------------------------------------------------------
* = Hyperthreading (2 = 1 Real + 1HT)
=============================================================================================
"gb_nuc" == Large Generalized Born Simulation. This is a large GB simulation with
25086 atoms, and is run with no cutoff and no rgbmax limit.
Shake is used and nrespa=1 (long range forces computed every step.)
Note: PMEMD 9.0 supports GB simulations so these timings are for PMEMD.
---------------------------------------------------------------------------------------------
name date CPU OS compiler npcu ps per day speedup
---------------------------------------------------------------------------------------------
Datastar 04/06 1.5GHz IBM PWR4 AIX XLF90 1 1.67 1.00
(SDSC) 2 3.23 1.93
(IBM Federation Switch) 4 6.12 3.65
8 11.33 6.77
16 22.97 13.74
32 45.04 26.93
64 89.19 53.33
128 170.99 102.24
192 248.80 148.76
256 322.85 193.03
384 459.97 275.02
512 573.55 342.94
768 745.60 445.81
1024 855.95 511.79
Bluegene/L 04/06 0.7GHz PowerPC Custom OS xlf90 1 0.34 1.00 (CO mode)
(SDSC) (440d) 2 0.68 2.02 (CO mode)
(custom torus and tree) (VN mode) 4 1.37 4.06
8 2.74 8.10
16 5.47 16.18
32 10.91 32.27
64 21.72 64.24
128 42.46 125.62
256 83.29 246.40
512 158.28 468.24
1024 286.66 848.05
---------------------------------------------------------------------------------------------
* = Hyperthreading (2 = 1 Real + 1HT)