Skip to content

NCCL benchmark: Bagua Net vs google fastsocket vs baseline

Xiangru Lian edited this page Sep 15, 2021 · 1 revision
baseline
#
# nThread 1 nGpus 1 minBytes 8 maxBytes 134217728 step: 2(factor) warmup iters: 5 iters: 20 validation: 1 
#
# Using devices
#   Rank  0 Pid 146066 on bjlt03-hi13 device  0 [0x1a] Tesla V100-SXM2-32GB
#   Rank  1 Pid  67084 on bjlt03-hi52 device  0 [0x1a] Tesla V100-SXM2-32GB
#   Rank  2 Pid  34899 on bjlt03-hi65 device  0 [0x1a] Tesla V100-SXM2-32GB
#   Rank  3 Pid  34368 on bjlt03-hi89 device  0 [0x1a] Tesla V100-SXM2-32GB
#
#                                                       out-of-place                       in-place          
#       size         count      type   redop     time   algbw   busbw  error     time   algbw   busbw  error
#        (B)    (elements)                       (us)  (GB/s)  (GB/s)            (us)  (GB/s)  (GB/s)       
           8             2     float     sum    300.4    0.00    0.00  1e-07    235.1    0.00    0.00  0e+00
          16             4     float     sum    245.7    0.00    0.00  3e-08    218.9    0.00    0.00  3e-08
          32             8     float     sum    195.7    0.00    0.00  3e-08    167.1    0.00    0.00  3e-08
          64            16     float     sum    151.3    0.00    0.00  3e-08    159.7    0.00    0.00  3e-08
         128            32     float     sum    158.0    0.00    0.00  3e-08    154.3    0.00    0.00  3e-08
         256            64     float     sum    154.2    0.00    0.00  3e-08    142.4    0.00    0.00  3e-08
         512           128     float     sum    144.7    0.00    0.01  3e-08    170.9    0.00    0.00  1e-08
        1024           256     float     sum    237.5    0.00    0.01  1e-07    260.9    0.00    0.01  1e-07
        2048           512     float     sum    357.5    0.01    0.01  1e-07    409.6    0.01    0.01  1e-07
        4096          1024     float     sum    376.4    0.01    0.02  2e-07    404.2    0.01    0.02  2e-07
        8192          2048     float     sum    421.2    0.02    0.03  2e-07    434.7    0.02    0.03  2e-07
       16384          4096     float     sum    421.4    0.04    0.06  2e-07    357.2    0.05    0.07  2e-07
       32768          8192     float     sum    443.5    0.07    0.11  2e-07    484.8    0.07    0.10  2e-07
       65536         16384     float     sum    494.6    0.13    0.20  2e-07    521.9    0.13    0.19  2e-07
      131072         32768     float     sum    530.4    0.25    0.37  2e-07    553.9    0.24    0.35  2e-07
      262144         65536     float     sum    645.5    0.41    0.61  2e-07    640.4    0.41    0.61  2e-07
      524288        131072     float     sum    584.3    0.90    1.35  2e-07    649.6    0.81    1.21  2e-07
     1048576        262144     float     sum    956.7    1.10    1.64  2e-07    972.9    1.08    1.62  2e-07
     2097152        524288     float     sum   1401.9    1.50    2.24  2e-07   1391.4    1.51    2.26  2e-07
     4194304       1048576     float     sum   2407.7    1.74    2.61  2e-07   2392.8    1.75    2.63  2e-07
     8388608       2097152     float     sum   4737.8    1.77    2.66  2e-07   4709.8    1.78    2.67  2e-07
    16777216       4194304     float     sum    10279    1.63    2.45  2e-07    10588    1.58    2.38  2e-07
    33554432       8388608     float     sum    20209    1.66    2.49  2e-07    20327    1.65    2.48  2e-07
    67108864      16777216     float     sum    40024    1.68    2.52  2e-07    39941    1.68    2.52  2e-07
   134217728      33554432     float     sum    80513    1.67    2.50  2e-07    81373    1.65    2.47  2e-07
# Out of bounds values : 0 OK
# Avg bus bandwidth    : 0.87012 
#

bagua-net
#
# nThread 1 nGpus 1 minBytes 8 maxBytes 134217728 step: 2(factor) warmup iters: 5 iters: 20 validation: 1 
#
# Using devices
#   Rank  0 Pid 146410 on bjlt03-hi13 device  0 [0x1a] Tesla V100-SXM2-32GB
#   Rank  1 Pid  67550 on bjlt03-hi52 device  0 [0x1a] Tesla V100-SXM2-32GB
#   Rank  2 Pid  35367 on bjlt03-hi65 device  0 [0x1a] Tesla V100-SXM2-32GB
#   Rank  3 Pid  34794 on bjlt03-hi89 device  0 [0x1a] Tesla V100-SXM2-32GB
#
#                                                       out-of-place                       in-place          
#       size         count      type   redop     time   algbw   busbw  error     time   algbw   busbw  error
#        (B)    (elements)                       (us)  (GB/s)  (GB/s)            (us)  (GB/s)  (GB/s)       
           8             2     float     sum    290.4    0.00    0.00  1e-07    299.9    0.00    0.00  0e+00
          16             4     float     sum    296.8    0.00    0.00  3e-08    297.2    0.00    0.00  3e-08
          32             8     float     sum    296.8    0.00    0.00  3e-08    298.1    0.00    0.00  3e-08
          64            16     float     sum    296.3    0.00    0.00  3e-08    295.5    0.00    0.00  3e-08
         128            32     float     sum    296.5    0.00    0.00  3e-08    296.3    0.00    0.00  3e-08
         256            64     float     sum    297.7    0.00    0.00  3e-08    296.1    0.00    0.00  3e-08
         512           128     float     sum    272.4    0.00    0.00  3e-08    286.4    0.00    0.00  1e-08
        1024           256     float     sum    287.8    0.00    0.01  1e-07    365.8    0.00    0.00  1e-07
        2048           512     float     sum    360.9    0.01    0.01  1e-07    330.6    0.01    0.01  1e-07
        4096          1024     float     sum    382.5    0.01    0.02  2e-07    400.0    0.01    0.02  2e-07
        8192          2048     float     sum    469.9    0.02    0.03  2e-07    524.6    0.02    0.02  2e-07
       16384          4096     float     sum    494.8    0.03    0.05  2e-07    488.9    0.03    0.05  2e-07
       32768          8192     float     sum    529.1    0.06    0.09  2e-07    523.0    0.06    0.09  2e-07
       65536         16384     float     sum    480.2    0.14    0.20  2e-07    467.3    0.14    0.21  2e-07
      131072         32768     float     sum    473.2    0.28    0.42  2e-07    474.4    0.28    0.41  2e-07
      262144         65536     float     sum    701.0    0.37    0.56  2e-07    727.9    0.36    0.54  2e-07
      524288        131072     float     sum    786.8    0.67    1.00  2e-07    821.5    0.64    0.96  2e-07
     1048576        262144     float     sum   1013.8    1.03    1.55  2e-07   1035.9    1.01    1.52  2e-07
     2097152        524288     float     sum   1139.9    1.84    2.76  2e-07   1112.1    1.89    2.83  2e-07
     4194304       1048576     float     sum   1731.0    2.42    3.63  2e-07   1779.1    2.36    3.54  2e-07
     8388608       2097152     float     sum   2725.0    3.08    4.62  2e-07   2915.8    2.88    4.32  2e-07
    16777216       4194304     float     sum   5586.3    3.00    4.50  2e-07   5632.8    2.98    4.47  2e-07
    33554432       8388608     float     sum    10860    3.09    4.63  2e-07    10412    3.22    4.83  2e-07
    67108864      16777216     float     sum    20490    3.28    4.91  2e-07    20323    3.30    4.95  2e-07
   134217728      33554432     float     sum    41678    3.22    4.83  2e-07    41141    3.26    4.89  2e-07
# Out of bounds values : 0 OK
# Avg bus bandwidth    : 1.35005 
#



google-fastsocket
#
# nThread 1 nGpus 1 minBytes 8 maxBytes 134217728 step: 2(factor) warmup iters: 5 iters: 20 validation: 1 
#
# Using devices
#   Rank  0 Pid 146098 on bjlt03-hi13 device  0 [0x1a] Tesla V100-SXM2-32GB
#   Rank  1 Pid  67132 on bjlt03-hi52 device  0 [0x1a] Tesla V100-SXM2-32GB
#   Rank  2 Pid  34947 on bjlt03-hi65 device  0 [0x1a] Tesla V100-SXM2-32GB
#   Rank  3 Pid  34416 on bjlt03-hi89 device  0 [0x1a] Tesla V100-SXM2-32GB
#
#                                                       out-of-place                       in-place          
#       size         count      type   redop     time   algbw   busbw  error     time   algbw   busbw  error
#        (B)    (elements)                       (us)  (GB/s)  (GB/s)            (us)  (GB/s)  (GB/s)       
           8             2     float     sum    201.8    0.00    0.00  1e-07    238.0    0.00    0.00  0e+00
          16             4     float     sum    155.5    0.00    0.00  3e-08    177.4    0.00    0.00  3e-08
          32             8     float     sum    161.2    0.00    0.00  3e-08    152.8    0.00    0.00  3e-08
          64            16     float     sum    177.1    0.00    0.00  3e-08    153.9    0.00    0.00  3e-08
         128            32     float     sum    159.9    0.00    0.00  3e-08    201.8    0.00    0.00  3e-08
         256            64     float     sum    201.7    0.00    0.00  3e-08    168.2    0.00    0.00  3e-08
         512           128     float     sum    169.1    0.00    0.00  3e-08    152.9    0.00    0.01  1e-08
        1024           256     float     sum    142.6    0.01    0.01  1e-07    145.5    0.01    0.01  1e-07
        2048           512     float     sum    176.8    0.01    0.02  1e-07    159.4    0.01    0.02  1e-07
        4096          1024     float     sum    151.4    0.03    0.04  2e-07    163.5    0.03    0.04  2e-07
        8192          2048     float     sum    189.8    0.04    0.06  2e-07    182.6    0.04    0.07  2e-07
       16384          4096     float     sum    202.3    0.08    0.12  2e-07    190.1    0.09    0.13  2e-07
       32768          8192     float     sum    277.7    0.12    0.18  2e-07    280.7    0.12    0.18  2e-07
       65536         16384     float     sum    329.9    0.20    0.30  2e-07    347.2    0.19    0.28  2e-07
      131072         32768     float     sum    402.3    0.33    0.49  2e-07    393.1    0.33    0.50  2e-07
      262144         65536     float     sum    470.2    0.56    0.84  2e-07    478.1    0.55    0.82  2e-07
      524288        131072     float     sum    523.4    1.00    1.50  2e-07    561.0    0.93    1.40  2e-07
     1048576        262144     float     sum    778.4    1.35    2.02  2e-07    844.8    1.24    1.86  2e-07
     2097152        524288     float     sum   1014.9    2.07    3.10  2e-07    977.6    2.15    3.22  2e-07
     4194304       1048576     float     sum   1501.6    2.79    4.19  2e-07   1549.8    2.71    4.06  2e-07
     8388608       2097152     float     sum    12722    0.66    0.99  2e-07   2922.5    2.87    4.31  2e-07
    16777216       4194304     float     sum   5016.7    3.34    5.02  2e-07   4994.6    3.36    5.04  2e-07
    33554432       8388608     float     sum   9430.3    3.56    5.34  2e-07   9842.1    3.41    5.11  2e-07
    67108864      16777216     float     sum    38740    1.73    2.60  2e-07    28907    2.32    3.48  2e-07
   134217728      33554432     float     sum    69368    1.93    2.90  2e-07    98598    1.36    2.04  2e-07
# Out of bounds values : 0 OK
# Avg bus bandwidth    : 1.24595
Clone this wiki locally