You are here: Home Resources Computers Historical Computers at PDC Lucidor II Memory bandwitdh

Memory bandwitdh

Memory bandwitdh results achieved by running the benchmark code stream2 on Lucidor.
The following results are from an adaption of the stream2 benchmark:
 Memory usage is         4882 Mbytes
 Smallest time delta is   9.536743164062500E-007
      Size  Iter     FILL      COPY     DAXPY       DOT
h15n22> ./stream2_alloc
 Memory usage is         4882 Mbytes
 Smallest time delta is   9.536743164062500E-007
      Size  Iter     FILL      COPY     DAXPY       DOT
        30    10   4645.4    9871.1   10683.2    7212.2      18.5
        51    10   7898.1   10073.8   12294.8    8261.9      18.5
        85    10  11453.2   12680.5   15024.5   10618.5      16.1
       144    10  13962.3   16774.1   22378.8   18072.0      11.6
       242    10  14682.3   17369.9   21198.0   17594.8       7.2
       408    10  17710.9   19149.0   25534.4   23292.7       5.2
       688    10  18833.1   19770.9   26365.3   24891.7       3.3
      1159    10  19175.7   20058.2   26026.9   24728.6       2.0
      1953    10  19964.1   20326.0   26675.1   25858.4       1.2
      3292    10  20320.1   20460.5   27188.7   26736.7       0.7
      5549    10  20466.1   20550.8   27315.7   27003.0       0.4
      9353    10  20580.7   20658.9   27450.9   27260.7       0.3
     15763    10  20624.8   19337.3   24230.2   21279.9       0.2
     26568    10  19700.9   14377.9   20691.7   15369.1       0.1
     44779    10  15255.5   13792.6   20598.0   16977.1       0.0
     75471    10  13884.2   13744.7   20979.3   15871.9       0.0
    127201    10  13826.0   10916.0   16417.2   13373.7       0.0
    214389    10  13776.0    4480.4    6845.7    8492.1       0.0
    361337    10   4627.1    3394.9    5114.4    6200.4       0.0
    609008    10   2570.8    3425.6    5159.3    6227.9       0.0
   1026439    10   2489.0    3432.2    5169.6    6239.3       0.0
   1729991    10   2487.8    3429.4    5170.3    6245.3       0.0
   2915779    10   2486.8    3426.6    5167.2    6244.9       0.0
   4914339    10   2485.6    3427.8    5165.3    6245.2       0.0
   8282773    10   2485.5    3427.0    5165.0    6246.3       0.0
  13960030    10   2483.5    3420.6    5150.6    6236.4       0.0
  23528647    10   2479.3    3407.1    5117.9    6212.9       0.0
  39655877    10   2484.3    3401.7    5105.3    6192.8       0.0
  66837187    10   2486.3    3398.6    5097.9    6183.3       0.0
 112649371    10   2487.5    3397.1    5093.8    6179.3       0.0
 189862578    10   2487.8    3395.2    5090.5    6176.0       0.0
 320000000    10   2486.7    3394.2    5088.1    6174.8       0.0
The numbers listed are million bytes/s. We used verion 10.0 of ifort and we compiled with
ifort -w -O3 -o stream2_alloc stream2_alloc.f second_wall_.o
where stream2_alloc.f is a version of stream2.f which uses allocatable array and has a correction of a bug in stream2.f. (The test called SUM in stream2.f is actually a DOTproduct and reads 16 Bytes/iter.) Note that we only used one CPU here.

The test was done 2008-05-16 on the node h15n22.