Main    
   
 
Home Contacts Sitemap ENG    UKR
 
 
Benchmarks

Tests of execution rates were taken from standard distributions of ATLAS and SCALAPACK.

Results of ATLAS testing on single processor:


LU factorisation tests

  • M,N - matrix dimensions;
  • TIME - CPU time in seconds;
  • MFLOP - execution rate;
  • RESID - residual of factorisation;
  • lda - leading matrix dimension.

complex single precision:

NREPS  Major      M      N    lda  NPVTS      TIME     MFLOP     RESID
=====  =====  =====  =====  =====  =====  ========  ========  ========
    0  Col      100    100    100     95     0.010   265.667 9.976e-02
    0  Col      200    200    200    193     0.030   709.778 7.859e-02
    0  Col      300    300    300    293     0.100   719.100 8.082e-02
    0  Col      400    400    400    396     0.190   897.404 7.025e-02
    0  Col      500    500    500    495     0.360   925.231 7.226e-02
    0  Col      600    600    600    592     0.560  1027.929 6.524e-02
    0  Col      700    700    700    698     0.870  1050.778 6.804e-02
    0  Col      800    800    800    797     1.220  1118.601 6.324e-02
    0  Col      900    900    900    895     1.640  1184.872 5.568e-02
    0  Col     1000   1000   1000    995     2.090  1275.439 5.835e-02

double precision 

NREPS  Major      M      N    lda  NPVTS      TIME     MFLOP     RESID
=====  =====  =====  =====  =====  =====  ========  ========  ========
    0  Col      100    100    100     95     0.000     0.000 2.033e-02
    0  Col      200    200    200    194     0.020   265.667 1.880e-02
    0  Col      300    300    300    295     0.050   359.100 1.366e-02
    0  Col      400    400    400    394     0.110   387.152 1.337e-02
    0  Col      500    500    500    490     0.200   416.042 1.226e-02
    0  Col      600    600    600    594     0.310   463.935 1.304e-02
    0  Col      700    700    700    693     0.520   439.272 1.045e-02
    0  Col      800    800    800    793     0.730   467.142 1.002e-02
    0  Col      900    900    900    893     1.040   466.918 1.117e-02
    0  Col     1000   1000   1000    995     1.410   472.459 9.980e-03

single precision

NREPS  Major      M      N    lda  NPVTS      TIME     MFLOP     RESID
=====  =====  =====  =====  =====  =====  ========  ========  ========
    0  Col      100    100    100     95     0.000     0.000 4.906e-02
    0  Col      200    200    200    194     0.010   531.333 3.685e-02
    0  Col      300    300    300    295     0.030   598.500 3.653e-02
    0  Col      400    400    400    394     0.050   851.733 3.627e-02
    0  Col      500    500    500    490     0.100   832.083 3.831e-02
    0  Col      600    600    600    594     0.190   756.947 3.060e-02
    0  Col      700    700    700    693     0.230   993.138 3.618e-02
    0  Col      800    800    800    793     0.330  1033.374 2.909e-02
    0  Col      900    900    900    893     0.440  1103.625 3.372e-02
    0  Col     1000   1000   1000    995     0.610  1092.077 3.269e-02

double complex precision

NREPS  Major      M      N    lda  NPVTS      TIME     MFLOP     RESID
=====  =====  =====  =====  =====  =====  ========  ========  ========
    0  Col      100    100    100     95     0.010   265.667 3.536e-02
    0  Col      200    200    200    193     0.050   425.867 2.882e-02
    0  Col      300    300    300    293     0.160   449.437 2.443e-02
    0  Col      400    400    400    396     0.340   501.490 2.266e-02
    0  Col      500    500    500    495     0.690   482.729 2.293e-02
    0  Col      600    600    600    592     1.170   492.000 2.234e-02
    0  Col      700    700    700    698     1.820   502.295 2.076e-02
    0  Col      800    800    800    797     2.580   528.951 1.922e-02
    0  Col      900    900    900    895     3.740   519.570 1.888e-02
    0  Col     1000   1000   1000    995     5.070   525.773 1.900e-02

NAS Parallel Benchmarks

The NAS Parallel Benchmarks (NPB) are a set of 8 programs designed to help evaluate the performance of parallel supercomputers:

  • FT - three-dimensional Fast Fourier Transform;
  • MG - multigrid solving of three-dimensional Poisson's equation;
  • LU, BT, SP - solving of finite difference discretization of the 3-D compressible Navier-Stokes equations by different factorization methods;
  • CG - solving of system of linear equations by conjugate gradient method;
  • EP - generation of the random numbers with Gaussian distribution;
  • IS - sorting of integer numbers.

Figure. 1. Cluster performance via number of running processes (size) for different benchmarks. Cluster performance with using of MOSIX system (index m) and without it.

Figure. 2. Running of processes with low priority (nice) for different benchmarks and different number of processes.


 
 
 
Kyiv National Taras Shevchenko University
© Information and Computer Center, 2002-2017