LINPACK

The original (1000x1000 double precision) LINPACK benchmark, with bugfixes, from

http://www.netlib.org/benchmark/1000d

It has been altered slightly to use cpu_time(t) instead of expecting seconds() to be defined, and Jack Dongarra's contact details are no longer printed with the results.

Build something like this (you may wish to add some compiler flags such as -O2):

gfortran -W -Wall 1000d.f -o linpack-benchmark

And run:

$ ./linpack-benchmark 
     norm. resid      resid           machep         x(1)          x(n)
  6.49150133E+00  7.20701276E-13  2.22044605E-16  1.00000000E+00  1.00000000E+00


    times are reported for matrices of order  1000
      factor     solve      total     mflops       unit      ratio
 times for array with leading dimension of1001
  1.028E+00  0.000E+00  1.028E+00  6.505E+02  3.075E-03  1.836E+01
  end of tests -- this version dated 10/12/92

This benchmark is here mainly for historical interest. For the benchmark used for top500 list submissions, you want the High Performance LINPACK (HPL) benchmark (and a suitable BLAS library): http://www.netlib.org/benchmark/hpl/