Personal tools
You are here: Home → Codes → ZEUS-MP → Benchmarks
Navigation
Log in


Forgot your password?
 
Document Actions

Benchmarks

by streeter — last modified 2007-03-30 04:05
Untitled Document
Summary

SGI/Cray Origin 2000
  • CODE: ZEUS-MP version 1.0
  • MACHINES: SGI/Cray Origin 2000
  • GEOMETRY: Cartesian XYZ
  • GRID: The physical grid is uniform and partitioned into 3-D "tiles" with variable (scale work) or fixed (fixed work) numbers of zones. Each process is assigned to 1 tile.
  • ALGORITHM: MPI used to pass messages; communication overlapped with computation; van Leer advection
  • PRECISION: DOUBLE PRECISION.
  • DATA: In the table below, "tused" is the average CPU seconds used by each process in computing the evolution (some system and ZEUS-MP overhead is excluded). The Zone-Cycles/sec is the total number of mesh zones times the number of time steps divided by tused.
  • PROBLEM TYPES:
    Hydrodynamics Benchmarks
    Magneto-Hydrodynamics Benchmarks
    Radiative Hydrodynamics Benchmarks
    Poisson Solvers Benchmarks
Hydrodynamics Benchmark
  • Problem: Blast -- the expansion of a hot sphere of plasma into an initially uniform medium (Sedov-Taylor Blast Wave):
  • ZeusMP version: 1.0
  • System configuration:
  • Machine: SGI Origin 2000 (R10000 195 Mhz CPUs)
  • OS version: SiliconGraphics IRIX6.5.4f
  • Compiling flags: ZEUS-MP is compiled with f77 -c -O3 -g3 -64 -mips4 -r10000 -OPT:roundoff=3,IEEE_arithmetic=3 
GRID: fixed 256 x 256 x 256 (30 steps)                         MFLOPS
 Processors  Layout  Wall Clock(s)  tused(s)  Zone-Cycles/sec  Speedup Speedup/Processor
      1       1x1x1   3.5313E+03   3.5112E+03    1.4309E+05      1.00        1.00
      2       2x1x1   1.8813E+03   1.8705E+03    2.6853E+05      1.88        0.94
      4       2x2x1   1.1321E+03   1.1254E+03    4.4627E+05      3.12        0.78
      8       2x2x2   5.1975E+02   5.1642E+02    9.7238E+05      6.80        0.85
     16       4x2x2   2.8300E+02   2.8098E+02    1.7863E+06     12.50        0.78
     32       4x4x2   1.4050E+02   1.3947E+02    3.5984E+06     25.18        0.79
     64       4x4x4   7.4563E+01   7.3919E+01    6.7854E+06     47.50        0.74
    128       8x4x4   5.2875E+01   5.1545E+01    9.7136E+06     68.12        0.53
    256(gsn)  8x8x4   2.5750E+01   2.5194E+01    1.9828E+07    139.37        0.54
    512(gsn)  8x8x8   1.7500E+01   1.7012E+01    2.8890E+07    206.40        0.40

GRID: fixed 512 x 512 x 512 (30 steps)                         MFLOPS
 Processors  Layout  Wall Clock(s)  tused(s)  Zone-Cycles/sec  Speedup Speedup/8 Proc.
      8       2x2x2   5.3834E+03   5.3524E+03    7.5229E+05      1.00       1.00
     16       4x2x2   2.7225E+03   2.7076E+03    1.4871E+06      1.98       0.99
     32       4x4x2   1.4439E+03   1.4345E+03    2.8070E+06      3.73       0.93
     64       4x4x4   5.5581E+02   5.5205E+02    7.2937E+06      9.70       1.21
    128       8x4x4   3.5080E+02   3.4651E+02    1.1620E+07      15.45      0.97
    256       8x8x4   2.0556E+02   2.0193E+02    1.9940E+07      26.19      0.82
    512(gsn)  8x8x8   1.3425E+02   1.3069E+02    3.0811E+07      40.96      0.64
Magneto-Hydrodynamics Benchmarks
  • Problem: Blast -- the expansion of a hot sphere of plasma into an initially uniform medium with uni-direction uniform magnetic field:
  • ZeusMP version: 1.0
  • System configuration:
  • Machine: SGI Origin 2000 (R10000 195 Mhz CPUs)
  • OS version: SiliconGraphics IRIX6.5.4f
  • Compiling flags: ZEUS-MP is compiled with f77 -c -O3 -g3 -64 -mips4 -r10000 -OPT:roundoff=3,IEEE_arithmetic=3 
GRID: fixed 256 x 256 x 256 (30 steps)                        MFLOPS
 Processors  Layout  Wall Clock(s)  tused(s)  Zone-Cycles/sec Speedup Speedup/8 Proc.
      8       2x2x2   2.4671E+03   2.4581E+03    2.0463E+05     1.00       1.00
     16       4x2x2   1.2191E+03   1.2140E+03    4.1418E+05     2.02       1.01
     32       4x4x2   5.8719E+02   5.8469E+02    8.5997E+05     4.20       1.05
     64       4x4x4   2.9969E+02   2.9805E+02    1.6859E+06     8.25       1.03
    128       8x4x4   1.5938E+02   1.5812E+02    3.1787E+06    15.55       0.97
    256(gsn)  8x8x4   7.3250E+01   7.2483E+01    6.9100E+06    33.91       1.06
    512(gsn)  8x8x8   6.0000E+01   5.9061E+01    8.4334E+06    41.62       0.65

GRID: fixed 512 x 512 x 512 (30 steps)                        MFLOPS
 Processors  Layout  Wall Clock(s)  tused(s)  Zone-Cycles/sec Speedup Speedup/16 Proc.
     16       4x2x2   1.1872E+04   1.1825E+04    3.4051E+05     1.00       1.00
     32       4x4x2   5.7961E+03   5.7703E+03    6.9781E+05     2.12       1.06
     64       4x4x4   2.9386E+03   2.9229E+03    1.3776E+06     4.05       1.01
    128       8x4x4   1.4267E+03   1.4141E+03    3.1787E+06     8.36       1.05
    256       8x8x4   7.7675E+02   7.7006E+02    5.2289E+06    15.36       0.96
    512(gsn)  8x8x8   4.3013E+02   4.2405E+02    9.4954E+06    27.89       0.87
Platinum IA-32 Linux Cluster
  • CODE: ZEUS-MP version 1.01
  • MACHINES: Platinum IA32 Linux Cluster
  • GEOMETRY: Cartesian XYZ
  • GRID: The physical grid is uniform and partitioned into 3-D "tiles" with variable (scale work) or fixed (fixed work) numbers of zones. Each process is assigned to 1 tile.
  • ALGORITHM: MPI used to pass messages; communication overlapped with computation; van Leer advection
  • PRECISION: DOUBLE PRECISION.
  • DATA: In the table below, "tused" is the average CPU seconds used by each process in computing the evolution (some system and ZEUS-MP overhead is excluded). The Zone-Cycles/sec is the total number of mesh zones times the number of time steps divided by tused.
  • PROBLEM TYPES:
    Hydrodynamics Benchmarks
    Magneto-Hydrodynamics Benchmarks
Hydrodynamics Benchmarks
  • Problem: Blast -- the expansion of a hot sphere of plasma into an initially uniform medium (Sedov-Taylor Blast Wave):
  • ZeusMP version: 1.01
  • System configuration:
  • Machine: Platinum IA32 Linux Cluster
  • OS version: Red Hat Linux release 6.2 Kernel 2.2.19smpx
  • Compiling flags: ZEUS-MP is compiled with ifc -c -O3 
GRID: fixed 256 x 256 x 256 (30 steps)                        MFLOPS
 Processors  Layout  Wall Clock(s)  tused(s)  Zone-Cycles/sec Speedup Speedup/8 Proc.
      8       2x2x2   4.2594E+02   4.4800E+02   1.1213E+06      1.00        1.00
     16       4x2x2   2.5666E+02   2.2625E+02   2.2198E+06      1.98        0.99
     32       4x4x2   1.1651E+02   1.2560E+02   3.9813E+06      3.57        0.89
     64       4x4x4   6.0999E+01   6.4960E+01   7.6972E+06      6.90        0.86
    128       8x4x4   2.4205E+01   3.7540E+01   1.3329E+07     11.93        0.75
    256       8x8x4   3.9606E+01   2.9120E+01   1.7149E+07     15.39        0.48

GRID: fixed 512 x 512 x 512 (30 steps)                        MFLOPS
 Processors  Layout  Wall Clock(s)  tused(s)  Zone-Cycles/sec Speedup Speedup/64 Proc.
     64       4x4x4   5.7435E+02   5.9830E+02   6.7300E+06      1.00        1.00
    128       8x4x4   3.2650E+02   2.9315E+02   1.3735E+07      2.04        1.02
    256       8x8x4   1.5479E+02   1.5067E+02   2.6724E+07      3.97        0.99
Magneto-Hydrodynamics Benchmarks
  • Problem: Blast -- the expansion of a hot sphere of plasma into an initially uniform medium with uni-direction uniform magnetic field:
  • ZeusMP version: 1.01
  • System configuration:
  • Machine: Platinum IA-32 Linux Cluster
  • OS version: Red Hat Linux release 6.2 Kernel 2.2.19smpx
  • Compiling flags: ZEUS-MP is compiled with ifc -c -O3 
GRID: fixed 256 x 256 x 256 (30 steps)                        MFLOPS
 Processors  Layout  Wall Clock(s)  tused(s)  Zone-Cycles/sec Speedup Speedup/8 Proc.
      8       2x2x2   1.2303E+03   1.2034E+03    4.1782E+05     1.00       1.00
     16       4x2x2   6.0194E+02   6.2282E+02    8.0728E+05     1.93       0.97
     32       4x4x2   3.4175E+02   3.3023E+02    1.5203E+06     3.64       0.91
     64       4x4x4   1.5862E+02   1.7122E+02    2.9283E+06     7.03       0.88
    128       8x4x4   9.8669E+01   9.0650E+01    5.5364E+06    13.28       0.83
    256       8x8x4   3.5096E+01   5.1570E+01    9.7034E+06    23.34       0.73

GRID: fixed 512 x 512 x 512 (30 steps)                        MFLOPS
 Processors  Layout  Wall Clock(s)  tused(s)  Zone-Cycles/sec Speedup Speedup/64 Proc.
     64       4x4x4   1.3812E+03   1.3694E+03    2.9404E+06     1.00       1.00
    128       8x4x4   7.2175E+02   7.1130E+02    5.6608E+06     1.93       0.96
    256       8x8x4   3.6330E+02   3.6265E+02    1.1103E+07     3.78       0.94
Titan IA-64 Linux Cluster
  • CODE: ZEUS-MP version 1.01
  • MACHINES: Titan IA64 Linux Cluster
  • GEOMETRY: Cartesian XYZ
  • GRID: The physical grid is uniform and partitioned into 3-D "tiles" with variable (scale work) or fixed (fixed work) numbers of zones. Each process is assigned to 1 tile.
  • ALGORITHM: MPI used to pass messages; communication overlapped with computation; van Leer advection
  • PRECISION: DOUBLE PRECISION.
  • DATA: In the table below, "tused" is the average CPU seconds used by each process in computing the evolution (some system and ZEUS-MP overhead is excluded). The Zone-Cycles/sec is the total number of mesh zones times the number of time steps divided by tused.
  • PROBLEM TYPES:
    Hydrodynamics Benchmarks
    Magneto-Hydrodynamics Benchmarks
Hydrodynamics Benchmarks
  • Problem: Blast -- the expansion of a hot sphere of plasma into an initially uniform medium (Sedov-Taylor Blast Wave):
  • ZeusMP version: 1.01
  • System configuration:
  • Machine: Titan IA64 Linux Cluster
  • OS version: Red Hat Linux release 7.1 Kernel 2.4.16
  • Compiling flags: ZEUS-MP is compiled with efc -c -O3 -ftz 
GRID: fixed 256 x 256 x 256 (30 steps)                        MFLOPS
 Processors  Layout  Wall Clock(s)  tused(s)  Zone-Cycles/sec Speedup Speedup/4 Proc.
      4       2x2x1   3.8400E+02   3.9503E+02   1.2719E+06      1.00        1.00
      8       2x2x2   1.9200E+02   2.1357E+02   2.3539E+06      1.85        0.93
     16       4x2x2   1.2800E+02   1.1508E+02   4.3683E+06      3.43        0.86
     32       4x4x2   5.6788E+01   5.6693E+01   8.8631E+06      6.97        0.87
     64       4x4x4   3.0597E+01   3.0480E+01   1.6450E+07     12.96        0.81
    128       8x4x4   1.6303E+01   1.6214E+01   3.0873E+07     24.36        0.76

GRID: fixed 512 x 512 x 512 (30 steps)                        MFLOPS
 Processors  Layout  Wall Clock(s)  tused(s)  Zone-Cycles/sec Speedup Speedup/32 Proc.
     32       4x4x2   5.1200E+02   5.2716E+02   7.6381E+06      1.00        1.00
     64       4x4x4   2.5600E+02   2.6164E+02   1.5390E+07      2.02        1.01
    128       8x4x4   1.2800E+02   1.2679E+02   3.1758E+07      4.16        1.04
Magneto-Hydrodynamics Benchmarks
  • Problem: Blast -- the expansion of a hot sphere of plasma into an initially uniform medium with uni-direction uniform magnetic field:
  • ZeusMP version: 1.01
  • System configuration:
  • Machine: Titan IA-64 Linux Cluster
  • OS version: Red Hat Linux release 7.1 Kernel 2.4.16
  • Compiling flags: ZEUS-MP is compiled with efc -c -O3 -ftz 
GRID: fixed 256 x 256 x 256 (30 steps)                         MFLOPS
 Processors  Layout  Wall Clock(s)  tused(s)  Zone-Cycles/sec  Speedup Speedup/4 Proc.
      4       2x2x1   1.2074E+03   1.2053E+03    4.1686E+05      1.00       1.00
      8       2x2x2   6.4000E+02   6.0782E+02    8.2762E+05      1.99       0.99
     16       4x2x2   3.2000E+02   3.2453E+02    1.5501E+06      3.71       0.93
     32       4x4x2   1.2800E+02   1.6390E+02    3.0656E+06      7.35       0.92
     64       4x4x4   6.4000E+01   8.8565E+01    5.6738E+06     13.61       0.85
    128       8x4x4   6.4000E+01   4.7618E+01    1.0562E+07     25.31       0.79

GRID: fixed 512 x 512 x 512 (30 steps)                        MFLOPS
 Processors  Layout  Wall Clock(s)  tused(s)  Zone-Cycles/sec Speedup Speedup/32 Proc.
     32       4x4x2   1.3440E+03   1.3258E+03    3.0371E+06     1.00       1.00
     64       4x4x4   7.0400E+02   6.8462E+02    5.8814E+06     1.94       0.97
    128       8x4x4   3.2000E+02   3.4359E+02    1.1719E+07     3.86       0.97
PSC Terascale Computing System
  • CODE: ZEUS-MP version 1.01
  • MACHINES: PSC Terascale Computing System (Compaq Alphaserver ES40)
  • GEOMETRY: Cartesian XYZ
  • GRID: The physical grid is uniform and partitioned into 3-D "tiles" with variable (scale work) or fixed (fixed work) numbers of zones. Each process is assigned to 1 tile.
  • ALGORITHM: MPI used to pass messages; communication overlapped with computation; van Leer advection
  • PRECISION: DOUBLE PRECISION.
  • DATA: In the table below, "tused" is the average CPU seconds used by each process in computing the evolution (some system and ZEUS-MP overhead is excluded). The Zone-Cycles/sec is the total number of mesh zones times the number of time steps divided by tused.
  • PROBLEM TYPES:
    Hydrodynamics Benchmarks
    Magneto-Hydrodynamics Benchmarks
Hydrodynamics Benchmarks
  • Problem: Blast -- the expansion of a hot sphere of plasma into an initially uniform medium (Sedov-Taylor Blast Wave):
  • ZeusMP version: 1.01
  • System configuration:
  • Machine: PSC Terascale Computing System (Compaq Alphaserver ES40)
  • OS version: Tru64 UNIX
  • Compiling flags: ZEUS-MP is compiled with f77 -c -O3 -OPT:roundoff=3,IEEE_arithmetic=3 
GRID: fixed 256 x 256 x 256 (30 steps)                        MFLOPS
 Processors  Layout  Wall Clock(s)  tused(s)  Zone-Cycles/sec Speedup Speedup/4 Proc.
      4       2x2x1   3.7998E+02   3.7859E+02   1.3262E+06      1.00        1.00
      8       2x2x2   1.9520E+02   1.9216E+02   2.6127E+06      1.97        0.99
     16       4x2x2   1.0344E+02   1.0132E+02   4.9560E+06      3.74        0.94
     32       4x4x2   5.4104E+01   5.0349E+01   9.9689E+06      7.52        0.94
     64       4x4x4   2.7884E+01   2.5834E+01   1.9421E+07     14.66        0.92
    128       8x4x4   1.8322E+01   1.4263E+01   3.5060E+07     26.54        0.83
    256       8x8x4   1.2815E+01   8.5420E+00   5.8702E+07     44.32        0.69
Magneto-Hydrodynamics Benchmarks
  • Problem: Blast -- the expansion of a hot sphere of plasma into an initially uniform medium with uni-direction uniform magnetic field:
  • ZeusMP version: 1.01
  • System configuration:
  • Machine: PSC Terascale Computing System (Compaq Alphaserver ES40)
  • OS version: Tru64 UNIX
  • Compiling flags: ZEUS-MP is compiled with f77 -c -O3 -OPT:roundoff=3,IEEE_arithmetic=3 
GRID: fixed 256 x 256 x 256 (30 steps)                        MFLOPS
 Processors  Layout  Wall Clock(s)  tused(s)  Zone-Cycles/sec Speedup Speedup/8 Proc.
      8       2x2x2   7.2296E+02   7.2033E+02    6.9811E+05     1.00       1.00
     16       4x2x2   3.7760E+02   3.7330E+02    1.3471E+06     1.93       0.97
     32       4x4x2   1.8024E+02   1.7666E+02    2.8458E+06     4.08       1.02
     64       4x4x4   8.6566E+01   8.3450E+01    6.0239E+06     8.63       1.08
    128       8x4x4   4.5811E+01   4.3601E+01    1.1525E+07    16.52       1.03
    256       8x8x4   2.5732E+01   2.3071E+01    2.1798E+07    31.22       0.98
Hydrodynamics:

Magneto-Hydrodynamics:
Poisson Solvers Benchmarks
  • Problem: Solving 3D gravitational potential of a uniform gas sphere with periodic boundary condition.
  • Poisson Solver: MGMPI, using multigrid method
  • System configuration:
    • Machine: SGI Origin 2000 (R10000 195 Mhz CPUs)
    • OS version: SiliconGraphics IRIX6.5.4f
    • Compiling flags: MGMPI is compiled with f77 -c -O3 -g3 -64 -mips4 -r10000 -OPT:roundoff=3,IEEE_arithmetic=3 
GRID: fixed 255 x 255 x 255
 Processors  Layout   CPU time(s)   Speedup   Speedup/8 Processors
      8       2x2x2     206.98        1.00          1.00
     16       4x2x2     112.02        1.85          0.93
     32       4x4x2      56.70        3.65          0.91
     64       4x4x4      56.43        3.66          0.46
    128       8x4x4      97.71        2.12          0.13
  • Problem: Solving 3D gravitational potential of a uniform gas sphere with periodic boundary condition.
  • Poisson Solver: FFTW, using fast Fourier transform method
  • System configuration:
    • Machine: SGI Origin 2000 (R10000 250 Mhz CPUs)
    • OS version: SiliconGraphics IRIX6.5.4f
    • Compiling flags: FFTW is compiled with f77 -c -O3 -g3 -64 -mips4 -r10000 -OPT:roundoff=3,IEEE_arithmetic=3 
GRID: fixed 256 x 256 x 256
  Processors  Layout   CPU time(s)   Speedup   Speedup/Processor
      1       1x1x1      40.72        1.00          1.00
      2       2x1x1      19.29        2.11          1.06
      4       4x1x1      10.83        3.76          0.94
      8       8x1x1       5.65        7.21          0.90
     16      16x1x1       2.47       16.49          1.03
     32      32x1x1       1.68       24.24          0.76
     64      64x1x1       3.28       12.41          0.19
    128     128x1x1       3.11       13.09          0.10


Back to ZEUS-MP 1.0
« August 2008 »
Su Mo Tu We Th Fr Sa
12
3456789
10111213141516
17181920212223
24252627282930
31
 
Powered by Plone CMS, the Open Source Content Management System

This site conforms to the following standards: