[Fwd: Benchmarking Beowulf]

[Jeff Candy]

Shu Xiao wrote:
 
> Any one knows what is the peak floating point performance for 
> Pentinum II? (in Mflop/s). Please help.

Well, "peak" performance in MFLOPS is equal to clock 
rate in MHZ.  Thus, a PII 450 cannot exceed 450 MFLOPS.
However, on linpack, I get:   

(1) Dual PII 350 (with 1 processor) gives 72.7 MFLOPS:

luna:/home/src/pgi/EXAMPLES/linpack/UNIX>setenv NCPUS 1
luna:/home/src/pgi/EXAMPLES/linpack/UNIX>linpkrd

     norm. resid      resid           machep         x(1)-1       
x(n)-1
  1.89264926E+00  8.39915160E-14  2.22044605E-16 -6.22835117E-14
-4.16333634E-14


    times are reported for matrices of order   100
      sgefa      sgesl      total     Kflops       unit      ratio
 times for array with leading dimension of 201
    0.00920    0.00030    0.00950     72288.    0.02767    0.16963
    0.00916    0.00029    0.00945     72655.    0.02753    0.16877
    0.00916    0.00029    0.00945     72670.    0.02752    0.16873
    0.00944    0.00000    0.00944     72713.    0.02751    0.16864

 times for array with leading dimension of 200
    0.00908    0.00029    0.00937     73275.    0.02729    0.16734
    0.00905    0.00029    0.00934     73511.    0.02721    0.16680
    0.00905    0.00029    0.00934     73518.    0.02720    0.16679
    0.00934    0.00000    0.00934     73536.    0.02720    0.16675

 ROLLED DOUBLE  PRECISION LINPACK PERFORMANCE        72713 KFLOPS 
FORTRAN STOP

(2) Dual PII 350 (with 2 processors) gives 119.3 MFLOPS (Portland 
    Group compiler with SMP support).

luna:/home/src/pgi/EXAMPLES/linpack/UNIX>setenv NCPUS 2
luna:/home/src/pgi/EXAMPLES/linpack/UNIX>linpkrd

     norm. resid      resid           machep         x(1)-1       
x(n)-1
  1.89264926E+00  8.39915160E-14  2.22044605E-16 -6.22835117E-14
-4.16333634E-14


    times are reported for matrices of order   100
      sgefa      sgesl      total     Kflops       unit      ratio
 times for array with leading dimension of 201
    0.00538    0.00045    0.00583    117720.    0.01699    0.10416
    0.00530    0.00045    0.00576    119316.    0.01676    0.10277
    0.00532    0.00044    0.00576    119213.    0.01678    0.10286
    0.00575    0.00000    0.00575    119354.    0.01676    0.10274

 times for array with leading dimension of 200
    0.00533    0.00044    0.00577    119068.    0.01680    0.10298
    0.00527    0.00043    0.00570    120363.    0.01662    0.10187
    0.00528    0.00043    0.00572    120131.    0.01665    0.10207
    0.00571    0.00000    0.00571    120326.    0.01662    0.10191

 ROLLED DOUBLE  PRECISION LINPACK PERFORMANCE       119354 KFLOPS 
FORTRAN STOP

(3) Single PII 450 gives 83.2 MFLOPS. 

athena:/u/candy/EXAMPLES/linpack/UNIX>linpkrd

     norm. resid      resid           machep         x(1)-1       
x(n)-1
  1.89264926E+00  8.39915160E-14  2.22044605E-16 -6.22835117E-14
-4.16333634E-14


    times are reported for matrices of order   100
      sgefa      sgesl      total     Kflops       unit      ratio
 times for array with leading dimension of 201
    0.00830    0.00024    0.00854     80378.    0.02488    0.15255
    0.02575    0.00100    0.02675     25671.    0.07791    0.47766
    0.00805    0.00024    0.00828     82901.    0.02413    0.14791
    0.00825    0.00000    0.00825     83182.    0.02404    0.14741

 times for array with leading dimension of 200
    0.02677    0.00036    0.02713     25306.    0.07903    0.48455
    0.00794    0.00024    0.00818     83985.    0.02381    0.14600
    0.02774    0.00026    0.02800     24527.    0.08154    0.49993
    0.00812    0.00000    0.00812     84553.    0.02365    0.14502

 ROLLED DOUBLE  PRECISION LINPACK PERFORMANCE        83182 KFLOPS 
FORTRAN STOP

The linpack benchmarks above are included with the Portland 
Group compiler.  Both machines have 256Mb of SDRAM.
 
-- 
Jeff Candy                       jcandy1@san.rr.com  

"Quite frankly, I'm against people who give vent to 
 their loquacity by extraneous, bombastic circum-
 locution" -- Eric Idle