Oneflow-Inc/OneAutoTest

202203_OneFlow_All_Models_Test_Report_On_master

github-actions opened this issue · 28 comments

202203_OneFlow_All_Models_Test_Report_On_master

表 2022-03-01 ResNet50 ddp dlperf 运行结果

0e930ab
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27348 MiB / 2336.08

表 2022-03-01 ResNet50 graph dlperf 运行结果

0e930ab
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24122 MiB / 9346.72
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24186 MiB / 3001.52
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24892 MiB / 10228.56
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24848 MiB / 3047.92

表 2022-03-01 ResNet50 lazy dlperf 运行结果

0e930ab
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24496 MiB / 10509.082
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23720 MiB / 10967.812
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24054 MiB / 2975.573
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23332 MiB / 3039.35

表 2022-03-01 WDL ddp dlperf 运行结果

0e930ab
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2804 MiB / 41.697254959493876
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2336 MiB / 6.381228551268578
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2574 MiB / 21.38000649958849

表 2022-03-01 WDL graph dlperf 运行结果

0e930ab
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2913 MiB / 21.38608274050057
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2131 MiB / 2.933517474681139
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2519 MiB / 12.059434475377202

表 2022-03-01 ResNet50 ddp dlperf 运行结果

b769add
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26916 MiB / 333.47

表 2022-03-01 ResNet50 graph dlperf 运行结果

b769add
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22984 MiB / 1408.41
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23072 MiB / 405.52
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23770 MiB / 1433.85
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23854 MiB / 401.37

表 2022-03-01 ResNet50 lazy dlperf 运行结果

b769add
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23878 MiB / 1442.12
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1442.363
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23426 MiB / 390.672
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22752 MiB / 392.874

表 2022-03-01 WDL ddp dlperf 运行结果

b769add
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5530 MiB / 301.34187519177794
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1922 MiB / 4.08247840963304
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3656 MiB / 120.96963089331985

表 2022-03-01 WDL graph dlperf 运行结果

b769add
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4404 MiB / 173.1843232549727
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1644 MiB / 1.5357423350214958
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3274 MiB / 94.99264749325812

表 2022-03-02 ResNet50 ddp dlperf 运行结果

0799967
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26916 MiB / 323.85

表 2022-03-02 ResNet50 ddp dlperf 运行结果

0799967
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27348 MiB / 2329.52

表 2022-03-02 ResNet50 graph dlperf 运行结果

0799967
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22984 MiB / 1361.54
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23128 MiB / 399.15
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23748 MiB / 1397.88
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23846 MiB / 394.52

表 2022-03-02 ResNet50 graph dlperf 运行结果

0799967
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24062 MiB / 9283.6
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24242 MiB / 2987.6
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24846 MiB / 10523.12
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24904 MiB / 3049.28

表 2022-03-02 ResNet50 lazy dlperf 运行结果

0799967
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23888 MiB / 1404.854
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1413.076
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23460 MiB / 382.429
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22782 MiB / 385.16

表 2022-03-02 ResNet50 lazy dlperf 运行结果

0799967
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24502 MiB / 10764.059
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23720 MiB / 11015.689
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24046 MiB / 2976.729
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23308 MiB / 3036.189

表 2022-03-02 WDL ddp dlperf 运行结果

0799967
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5530 MiB / 303.03840183094144
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1922 MiB / 4.080690834671259
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3656 MiB / 121.4067801758647

表 2022-03-02 WDL graph dlperf 运行结果

0799967
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4914 MiB / 175.71181580796838
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1644 MiB / 1.5919778645038605
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3274 MiB / 98.59595589339733

表 2022-03-02 WDL ddp dlperf 运行结果

bc2a158
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2804 MiB / 41.33593804016709
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2336 MiB / 6.503182040527463
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2574 MiB / 21.402303118258715

表 2022-03-02 WDL graph dlperf 运行结果

bc2a158
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2849 MiB / 21.359103625640273
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2131 MiB / 3.01633065007627
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2519 MiB / 12.284300798550248

表 2022-03-03 ResNet50 ddp dlperf 运行结果

59b40b9
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26916 MiB / 330.55

表 2022-03-03 ResNet50 ddp dlperf 运行结果

59b40b9
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27348 MiB / 2333.84

表 2022-03-03 ResNet50 graph dlperf 运行结果

59b40b9
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22984 MiB / 1408.04
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23116 MiB / 405.97
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23762 MiB / 1434.9
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23874 MiB / 404.51

表 2022-03-03 ResNet50 graph dlperf 运行结果

59b40b9
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24082 MiB / 5594.8
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24214 MiB / 2968.16
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g None MiB / 0
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g None MiB / 0

表 2022-03-03 ResNet50 lazy dlperf 运行结果

59b40b9
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23928 MiB / 1443.449
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1435.946
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23450 MiB / 390.786
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22776 MiB / 392.839

表 2022-03-03 ResNet50 lazy dlperf 运行结果

59b40b9
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24514 MiB / 10594.144
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23720 MiB / 11073.852
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24004 MiB / 2991.026
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23260 MiB / 3038.091

表 2022-03-03 WDL ddp dlperf 运行结果

59b40b9
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5530 MiB / 302.41639285720885
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1922 MiB / 4.074203174561262
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3656 MiB / 115.92605589702725

表 2022-03-03 WDL graph dlperf 运行结果

59b40b9
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4404 MiB / 178.33482718467712
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1644 MiB / 1.5946020893752575
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3018 MiB / 99.99703418277204

表 2022-03-04 ResNet50 ddp dlperf 运行结果

f98e143
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26918 MiB / 329.49

表 2022-03-04 ResNet50 ddp dlperf 运行结果

f98e143
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27350 MiB / 2328.56

表 2022-03-04 ResNet50 graph dlperf 运行结果

f98e143
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22986 MiB / 1406.09
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23142 MiB / 405.78
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23750 MiB / 1416.14
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23802 MiB / 402.67

表 2022-03-04 ResNet50 graph dlperf 运行结果

f98e143
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24084 MiB / 5428.16
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24206 MiB / 2945.04
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g None MiB / 0
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g None MiB / 0

表 2022-03-04 ResNet50 lazy dlperf 运行结果

f98e143
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23902 MiB / 1434.54
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1418.704
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23424 MiB / 390.4
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22748 MiB / 393.389

表 2022-03-04 ResNet50 lazy dlperf 运行结果

f98e143
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24502 MiB / 10422.073
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23720 MiB / 11034.553
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 23974 MiB / 3003.858
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23272 MiB / 3031.699

表 2022-03-04 WDL ddp dlperf 运行结果

f98e143
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5532 MiB / 302.3173070643097
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1924 MiB / 4.122759714722633
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3658 MiB / 121.65466794371605

表 2022-03-04 WDL ddp dlperf 运行结果

f98e143
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2806 MiB / 41.352347280830145
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2338 MiB / 6.4914287980645895
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2576 MiB / 21.71202007867396

表 2022-03-04 WDL graph dlperf 运行结果

f98e143
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4916 MiB / 175.94144348055124
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1648 MiB / 1.4607818201184273
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3276 MiB / 95.0873240083456

表 2022-03-04 WDL graph dlperf 运行结果

f98e143
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2849 MiB / 32.70790943130851
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2133 MiB / 3.20513266697526
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2519 MiB / 18.708423878997564

表 2022-03-05 ResNet50 ddp dlperf 运行结果

952e06f
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26918 MiB / 325.23

表 2022-03-05 ResNet50 ddp dlperf 运行结果

952e06f
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27350 MiB / 2326.32

表 2022-03-05 ResNet50 graph dlperf 运行结果

952e06f
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22986 MiB / 1383.28
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23144 MiB / 400.59
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23744 MiB / 1413.1
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23838 MiB / 399.05

表 2022-03-05 ResNet50 graph dlperf 运行结果

952e06f
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24084 MiB / 5394.88
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24196 MiB / 2913.84
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g None MiB / 0
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g None MiB / 0

表 2022-03-05 ResNet50 lazy dlperf 运行结果

952e06f
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23918 MiB / 1432.202
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1444.211
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23400 MiB / 388.682
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22756 MiB / 391.443

表 2022-03-05 ResNet50 lazy dlperf 运行结果

952e06f
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24490 MiB / 10680.286
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23722 MiB / 10977.762
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24002 MiB / 3000.611
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23260 MiB / 3036.545

表 2022-03-05 WDL ddp dlperf 运行结果

952e06f
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5532 MiB / 300.93262688815594
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1924 MiB / 4.056669542565942
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3658 MiB / 119.27725131995976

表 2022-03-05 WDL ddp dlperf 运行结果

952e06f
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2806 MiB / 40.70089875906706
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2338 MiB / 6.412878097966313
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2576 MiB / 20.94587479904294

表 2022-03-05 WDL graph dlperf 运行结果

952e06f
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4406 MiB / 173.8796822205186
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1646 MiB / 1.5672107730060816
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3276 MiB / 93.7170257717371

表 2022-03-05 WDL graph dlperf 运行结果

952e06f
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2897 MiB / 34.413256427273154
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2133 MiB / 3.1653102841228247
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2487 MiB / 17.743708645924926

表 2022-03-06 ResNet50 ddp train 50E运行结果

9a86377
Memory/Throughput/TOP1
ResNet50_ddp_train_cpudecode_FP32_b192_1n8g 20784 MiB / 2287.04 / 76.32

表 2022-03-06 ResNet50 graph train 50E运行结果

9a86377
Memory/Throughput/TOP1
ResNet50_graph_train_gpudecode_FP16_b192_1n8g 11486 MiB / 10040.08 / 76.192

表 2022-03-06 ResNet50 graph train 50E运行结果

9a86377
Memory/Throughput/TOP1
ResNet50_graph_train_gpudecode_FP16_b192_1n4g 10922 MiB / 4856.48 / 75.98799999999999

表 2022-03-06 ResNet50 ddp dlperf 运行结果

afd568f
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26918 MiB / 323.47

表 2022-03-06 ResNet50 ddp dlperf 运行结果

afd568f
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27350 MiB / 2319.84

表 2022-03-06 ResNet50 graph dlperf 运行结果

afd568f
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22986 MiB / 1372.34
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23080 MiB / 396.48
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23742 MiB / 1408.5
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23816 MiB / 394.66

表 2022-03-06 ResNet50 graph dlperf 运行结果

afd568f
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24066 MiB / 8669.12
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24134 MiB / 2986.64
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24892 MiB / 10628.8
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24924 MiB / 2991.28

表 2022-03-06 ResNet50 lazy dlperf 运行结果

afd568f
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23894 MiB / 1406.552
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1440.649
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23426 MiB / 381.234
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22698 MiB / 383.431

表 2022-03-06 ResNet50 lazy dlperf 运行结果

afd568f
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24484 MiB / 10621.911
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23720 MiB / 10963.874
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24032 MiB / 2968.907
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23284 MiB / 3034.997

表 2022-03-06 WDL ddp dlperf 运行结果

afd568f
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5532 MiB / 301.00631874985993
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1924 MiB / 4.090983739122748
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3658 MiB / 119.79926997609437

表 2022-03-06 WDL ddp dlperf 运行结果

afd568f
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2806 MiB / 41.51570396870375
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2338 MiB / 6.492016104981303
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2576 MiB / 21.50454654544592

表 2022-03-06 WDL graph dlperf 运行结果

afd568f
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4916 MiB / 175.5982470177114
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1646 MiB / 1.5786698088049889
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3212 MiB / 99.78619939647615

表 2022-03-06 WDL graph dlperf 运行结果

afd568f
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2915 MiB / 21.88727445155382
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2133 MiB / 2.9868613854050636
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2521 MiB / 12.351355291903019

表 2022-03-07 ResNet50 ddp dlperf 运行结果

31bb9f1
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26922 MiB / 327.54

表 2022-03-07 ResNet50 ddp dlperf 运行结果

31bb9f1
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27356 MiB / 2327.2

表 2022-03-07 ResNet50 graph dlperf 运行结果

31bb9f1
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22992 MiB / 1391.94
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23078 MiB / 400.62
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23778 MiB / 1419.23
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23810 MiB / 398.97

表 2022-03-07 ResNet50 graph dlperf 运行结果

31bb9f1
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24068 MiB / 9366.32
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24140 MiB / 3002.72
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24844 MiB / 10188.64
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24944 MiB / 3046.56

表 2022-03-07 ResNet50 lazy dlperf 运行结果

31bb9f1
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23922 MiB / 1423.063
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1423.954
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23454 MiB / 387.57
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22756 MiB / 391.363

表 2022-03-07 ResNet50 lazy dlperf 运行结果

31bb9f1
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24494 MiB / 10670.743
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23722 MiB / 10952.26
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24042 MiB / 2944.978
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23280 MiB / 3033.521

表 2022-03-07 WDL ddp dlperf 运行结果

31bb9f1
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5538 MiB / 300.88094582408667
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1930 MiB / 3.924751177430153
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3664 MiB / 122.6026470810175

表 2022-03-07 WDL ddp dlperf 运行结果

31bb9f1
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2812 MiB / 40.89517593383789
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2344 MiB / 6.498464183881879
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2582 MiB / 21.620086017996073

表 2022-03-07 WDL graph dlperf 运行结果

31bb9f1
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g None MiB / 0

表 2022-03-07 WDL graph dlperf 运行结果

31bb9f1
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g None MiB / 0

表 2022-03-07 ResNet50 ddp train 50E运行结果

9a86377
Memory/Throughput/TOP1
ResNet50_ddp_train_cpudecode_FP32_b192_1n4g 20420 MiB / 1222.88 / 76.27199999999999

表 2022-03-08 ResNet50 ddp dlperf 运行结果

687dcdd
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26934 MiB / 325.01

表 2022-03-08 ResNet50 ddp dlperf 运行结果

687dcdd
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27368 MiB / 2321.76

表 2022-03-08 ResNet50 graph dlperf 运行结果

687dcdd
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23004 MiB / 1402.75
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23102 MiB / 400.94
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23774 MiB / 1426.55
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23814 MiB / 399.66

表 2022-03-08 ResNet50 graph dlperf 运行结果

687dcdd
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24100 MiB / 9446.32
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24212 MiB / 3015.92
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24918 MiB / 10508.16
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24930 MiB / 3043.84

表 2022-03-08 ResNet50 lazy dlperf 运行结果

687dcdd
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23902 MiB / 1417.199
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1429.309
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23454 MiB / 390.584
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22696 MiB / 390.194

表 2022-03-08 ResNet50 lazy dlperf 运行结果

687dcdd
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24494 MiB / 10641.573
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23722 MiB / 10964.334
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24030 MiB / 3012.314
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23280 MiB / 3038.909

表 2022-03-08 WDL ddp dlperf 运行结果

687dcdd
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5550 MiB / 302.8996437191963
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1942 MiB / 4.0749239306896925
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3676 MiB / 116.87632458284497

表 2022-03-08 WDL ddp dlperf 运行结果

687dcdd
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2824 MiB / 41.753738697618246
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2356 MiB / 6.557452050969005
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2594 MiB / 21.556252690032125

表 2022-03-08 WDL graph dlperf 运行结果

687dcdd
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g None MiB / 0

表 2022-03-08 WDL graph dlperf 运行结果

687dcdd
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g None MiB / 0

表 2022-03-09 ResNet50 ddp dlperf 运行结果

f2c4d41
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26936 MiB / 327.71

表 2022-03-09 ResNet50 ddp dlperf 运行结果

f2c4d41
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27368 MiB / 2310.64

表 2022-03-09 ResNet50 graph dlperf 运行结果

f2c4d41
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23004 MiB / 1416.71
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23094 MiB / 405.0
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23784 MiB / 1429.49
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23876 MiB / 402.17

表 2022-03-09 ResNet50 graph dlperf 运行结果

f2c4d41
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24080 MiB / 9645.12
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24158 MiB / 3078.08
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24896 MiB / 10554.16
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24964 MiB / 3043.52

表 2022-03-09 ResNet50 lazy dlperf 运行结果

f2c4d41
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23882 MiB / 1427.933
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1440.353
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23480 MiB / 389.653
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22696 MiB / 392.944

表 2022-03-09 ResNet50 lazy dlperf 运行结果

f2c4d41
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24486 MiB / 10518.548
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23722 MiB / 10926.907
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24096 MiB / 3005.384
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23270 MiB / 3038.458

表 2022-03-09 WDL ddp dlperf 运行结果

f2c4d41
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5550 MiB / 301.92956932261586
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1942 MiB / 4.049612004309893
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3676 MiB / 119.31868255883455

表 2022-03-09 WDL ddp dlperf 运行结果

f2c4d41
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2824 MiB / 42.14073222503066
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2356 MiB / 6.642493102699518
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2594 MiB / 21.95678086951375

表 2022-03-09 WDL graph dlperf 运行结果

f2c4d41
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g None MiB / 0

表 2022-03-09 WDL graph dlperf 运行结果

f2c4d41
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g None MiB / 0

表 2022-03-10 ResNet50 ddp dlperf 运行结果

a81140c
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26936 MiB / 349.07

表 2022-03-10 ResNet50 ddp dlperf 运行结果

a81140c
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27368 MiB / 2516.72

表 2022-03-10 ResNet50 graph dlperf 运行结果

a81140c
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23002 MiB / 1462.3
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23094 MiB / 396.58
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23792 MiB / 1415.19
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23886 MiB / 396.23

表 2022-03-10 ResNet50 graph dlperf 运行结果

a81140c
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24080 MiB / 10368.4
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24194 MiB / 3106.16
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24892 MiB / 10606.56
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24922 MiB / 3046.4

表 2022-03-10 ResNet50 lazy dlperf 运行结果

a81140c
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23912 MiB / 1425.776
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1422.741
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23448 MiB / 384.063
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22748 MiB / 385.899

表 2022-03-10 ResNet50 lazy dlperf 运行结果

a81140c
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24520 MiB / 10590.304
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23722 MiB / 10912.442
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24042 MiB / 2994.398
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23284 MiB / 3038.379

表 2022-03-10 WDL ddp dlperf 运行结果

a81140c
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5550 MiB / 301.10769851878285
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1942 MiB / 4.0765736773610115
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3676 MiB / 118.49302869290113

表 2022-03-10 WDL ddp dlperf 运行结果

a81140c
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2824 MiB / 41.825400829315186
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2356 MiB / 6.470005005598068
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2594 MiB / 21.202367953956127

表 2022-03-10 WDL graph dlperf 运行结果

a81140c
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g None MiB / 0

表 2022-03-10 WDL graph dlperf 运行结果

a81140c
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g None MiB / 0

表 2022-03-11 WDL ddp dlperf 运行结果

277e7a3
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5388 MiB / 301.61528332158923
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1798 MiB / 4.073721908032894
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3514 MiB / 115.3332638591528

表 2022-03-11 WDL ddp dlperf 运行结果

277e7a3
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2662 MiB / 42.54867171496153
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2212 MiB / 6.4938247166574
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2432 MiB / 21.24959811195731

表 2022-03-11 WDL graph dlperf 运行结果

277e7a3
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g None MiB / 0

表 2022-03-11 WDL graph dlperf 运行结果

277e7a3
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g None MiB / 0

表 2022-03-12 ResNet50 ddp dlperf 运行结果

bada333
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26936 MiB / 346.02

表 2022-03-12 ResNet50 ddp dlperf 运行结果

bada333
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27368 MiB / 2494.8

表 2022-03-12 ResNet50 graph dlperf 运行结果

bada333
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23004 MiB / 1479.17
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23148 MiB / 398.85
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23806 MiB / 1427.24
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23814 MiB / 395.0

表 2022-03-12 ResNet50 graph dlperf 运行结果

bada333
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24122 MiB / 10421.04
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24192 MiB / 3112.96
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24912 MiB / 10760.0
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24968 MiB / 3066.8

表 2022-03-12 ResNet50 lazy dlperf 运行结果

bada333
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23912 MiB / 1412.691
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1415.157
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23468 MiB / 384.089
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22772 MiB / 386.615

表 2022-03-12 ResNet50 lazy dlperf 运行结果

bada333
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24510 MiB / 10727.877
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23722 MiB / 10765.699
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24054 MiB / 3001.736
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23318 MiB / 3031.407

表 2022-03-12 WDL ddp dlperf 运行结果

bada333
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5388 MiB / 301.81650764867663
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1798 MiB / 4.083905391395092
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3514 MiB / 120.04651974141598

表 2022-03-12 WDL ddp dlperf 运行结果

bada333
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2662 MiB / 41.39880060404539
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2212 MiB / 6.473604127764702
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2432 MiB / 21.689473174512386

表 2022-03-12 WDL graph dlperf 运行结果

bada333
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g None MiB / 0

表 2022-03-12 WDL graph dlperf 运行结果

bada333
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g None MiB / 0

表 2022-03-13 ResNet50 ddp dlperf 运行结果

1c48559
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 347.82

表 2022-03-13 ResNet50 ddp dlperf 运行结果

1c48559
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2530.4

表 2022-03-13 ResNet50 graph dlperf 运行结果

1c48559
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1487.51
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23134 MiB / 400.3
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23772 MiB / 1419.66
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23874 MiB / 398.73

表 2022-03-13 ResNet50 graph dlperf 运行结果

1c48559
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24126 MiB / 10575.04
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24214 MiB / 3105.76
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24908 MiB / 10759.68
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24892 MiB / 3060.16

表 2022-03-13 ResNet50 lazy dlperf 运行结果

1c48559
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23928 MiB / 1420.126
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1431.12
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23470 MiB / 387.507
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22748 MiB / 390.584

表 2022-03-13 ResNet50 lazy dlperf 运行结果

1c48559
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24500 MiB / 10441.406
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23722 MiB / 11010.022
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24032 MiB / 3004.745
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23258 MiB / 3035.369

表 2022-03-13 WDL ddp dlperf 运行结果

1c48559
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5390 MiB / 303.95628552883863
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1800 MiB / 4.074000261723995
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3516 MiB / 117.5428616963327

表 2022-03-13 WDL ddp dlperf 运行结果

1c48559
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2664 MiB / 40.89858549460769
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2214 MiB / 6.490217313170433
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2434 MiB / 21.287184588611126

表 2022-03-13 WDL graph dlperf 运行结果

1c48559
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g None MiB / 0

表 2022-03-13 WDL graph dlperf 运行结果

1c48559
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g None MiB / 0

表 2022-03-13 ResNet50 ddp train 50E运行结果

277e7a3
Memory/Throughput/TOP1
ResNet50_ddp_train_cpudecode_FP32_b192_1n8g 20802 MiB / 2465.04 / 76.292

表 2022-03-13 ResNet50 graph train 50E运行结果

277e7a3
Memory/Throughput/TOP1
ResNet50_graph_train_gpudecode_FP16_b192_1n8g 11464 MiB / 10389.52 / 76.07

表 2022-03-13 ResNet50 graph train 50E运行结果

277e7a3
Memory/Throughput/TOP1
ResNet50_graph_train_gpudecode_FP16_b192_1n4g 10803 MiB / 5312.84 / 76.112

表 2022-03-14 ResNet50 ddp train 50E运行结果

277e7a3
Memory/Throughput/TOP1
ResNet50_ddp_train_cpudecode_FP32_b192_1n4g 20490 MiB / 1280.76 / 76.196

表 2022-03-14 ResNet50 graph train 50E运行结果

277e7a3
Memory/Throughput/TOP1
ResNet50_graph_train_gpudecode_FP16_b192_1n4g 10740 MiB / 5141.16 / 76.164

表 2022-03-14 ResNet50 ddp dlperf 运行结果

e3a35be
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 344.87

表 2022-03-14 ResNet50 graph dlperf 运行结果

e3a35be
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1467.12
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23098 MiB / 395.07
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23796 MiB / 1389.13
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23904 MiB / 394.13

表 2022-03-14 ResNet50 lazy dlperf 运行结果

e3a35be
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23928 MiB / 1390.417
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1428.364
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23482 MiB / 379.291
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22772 MiB / 381.054

表 2022-03-14 WDL ddp dlperf 运行结果

e3a35be
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5390 MiB / 305.7424729205668
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1800 MiB / 4.301084879785776
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3516 MiB / 127.91095750033855

表 2022-03-14 WDL graph dlperf 运行结果

e3a35be
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4266 MiB / 158.28211283683777
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1508 MiB / 1.4573910012841225
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3136 MiB / 84.5969534702599

表 2022-03-15 ResNet50 ddp dlperf 运行结果

927e05b
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 355.56

表 2022-03-15 ResNet50 ddp dlperf 运行结果

927e05b
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2506.32

表 2022-03-15 ResNet50 graph dlperf 运行结果

927e05b
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1473.46
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23150 MiB / 405.67
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23794 MiB / 1438.79
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23810 MiB / 402.35

表 2022-03-15 ResNet50 graph dlperf 运行结果

927e05b
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24124 MiB / 10478.32
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24196 MiB / 3112.64
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24914 MiB / 10262.8
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24900 MiB / 3054.08

表 2022-03-15 ResNet50 lazy dlperf 运行结果

927e05b
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23878 MiB / 1433.078
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1449.743
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23468 MiB / 391.018
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22720 MiB / 392.646

表 2022-03-15 ResNet50 lazy dlperf 运行结果

927e05b
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24500 MiB / 10431.883
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23720 MiB / 10825.055
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24040 MiB / 2985.043
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23322 MiB / 3033.444

表 2022-03-15 WDL ddp dlperf 运行结果

927e05b
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5390 MiB / 302.854912687093
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1800 MiB / 4.082240626215935
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3516 MiB / 118.30731334537268

表 2022-03-15 WDL ddp dlperf 运行结果

927e05b
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2664 MiB / 42.01752261444926
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2214 MiB / 6.475165244191885
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2434 MiB / 21.52884555980563

表 2022-03-15 WDL graph dlperf 运行结果

927e05b
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4266 MiB / 173.9313365444541
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1508 MiB / 1.5834473744034767
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3072 MiB / 98.79627230018377

表 2022-03-15 WDL graph dlperf 运行结果

927e05b
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2917 MiB / 21.65032197907567
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2135 MiB / 2.7416682951152325
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2491 MiB / 12.386275254189968

表 2022-03-16 ResNet50 ddp dlperf 运行结果

09bbafe
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2518.96

表 2022-03-16 ResNet50 graph dlperf 运行结果

09bbafe
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24084 MiB / 10547.76
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24174 MiB / 3109.84
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24900 MiB / 10259.52
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24990 MiB / 3043.6

表 2022-03-16 ResNet50 lazy dlperf 运行结果

09bbafe
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24502 MiB / 11000.238
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23722 MiB / 10989.743
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24006 MiB / 2995.667
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23304 MiB / 3022.653

表 2022-03-16 ResNet50 ddp dlperf 运行结果

66e1957
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 346.87

表 2022-03-16 ResNet50 graph dlperf 运行结果

66e1957
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1479.85
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23102 MiB / 395.2
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23800 MiB / 1408.89
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23884 MiB / 395.7

表 2022-03-16 ResNet50 lazy dlperf 运行结果

66e1957
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23896 MiB / 1411.582
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1407.69
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23482 MiB / 380.636
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22698 MiB / 382.814

表 2022-03-16 WDL ddp dlperf 运行结果

66e1957
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5390 MiB / 302.52059254795313
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1800 MiB / 3.921513345092535
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3516 MiB / 117.96464419737458

表 2022-03-16 WDL ddp dlperf 运行结果

66e1957
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2664 MiB / 40.75848162174225
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2214 MiB / 6.550648972392082
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2434 MiB / 21.67178449034691

表 2022-03-16 WDL graph dlperf 运行结果

66e1957
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4776 MiB / 174.49466932192445
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1506 MiB / 1.4670290015637875
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3136 MiB / 96.29220212250948

表 2022-03-16 WDL graph dlperf 运行结果

66e1957
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2901 MiB / 21.699180584400892
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2135 MiB / 2.7533767111599445
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2491 MiB / 12.12074588239193

表 2022-03-17 ResNet50 ddp dlperf 运行结果

70270a21e
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 347.38

表 2022-03-17 ResNet50 ddp dlperf 运行结果

70270a21e
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2515.76

表 2022-03-17 ResNet50 graph dlperf 运行结果

70270a21e
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1455.25
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23094 MiB / 398.52
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23816 MiB / 1417.92
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23880 MiB / 397.15

表 2022-03-17 ResNet50 graph dlperf 运行结果

70270a21e
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24146 MiB / 10402.48
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24268 MiB / 3111.44
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24904 MiB / 10730.72
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24892 MiB / 3056.16

表 2022-03-17 ResNet50 lazy dlperf 运行结果

70270a21e
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23888 MiB / 1413.618
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1413.811
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23462 MiB / 381.855
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22748 MiB / 385.076

表 2022-03-17 ResNet50 lazy dlperf 运行结果

70270a21e
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24540 MiB / 10719.885
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23722 MiB / 10925.244
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24044 MiB / 2986.521
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23308 MiB / 3035.195

表 2022-03-17 WDL ddp dlperf 运行结果

70270a21e
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5390 MiB / 302.1350906044245
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1800 MiB / 4.072536181658506
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3516 MiB / 120.72668739035726

表 2022-03-17 WDL ddp dlperf 运行结果

70270a21e
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2664 MiB / 42.24585598707199
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2214 MiB / 6.514367386698723
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2434 MiB / 20.5202641710639

表 2022-03-17 WDL graph dlperf 运行结果

70270a21e
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4266 MiB / 173.51090018451214
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1506 MiB / 1.5623415522277355
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 2880 MiB / 91.30167655646801

表 2022-03-17 WDL graph dlperf 运行结果

70270a21e
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2853 MiB / 21.836467061191797
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2135 MiB / 2.7435242980718613
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2523 MiB / 12.311562813818455

表 2022-03-17 ResNet50 ddp dlperf 运行结果

a33a825
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 348.15

表 2022-03-17 ResNet50 ddp dlperf 运行结果

a33a825
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2526.88

表 2022-03-17 ResNet50 ddp train 50E运行结果

a33a825
Memory/Throughput/TOP1
ResNet50_ddp_train_cpudecode_FP32_b192_1n8g 20804 MiB / 2468.64 / 76.14999999999999

表 2022-03-17 ResNet50 ddp train 50E运行结果

a33a825
Memory/Throughput/TOP1
ResNet50_ddp_train_cpudecode_FP32_b192_1n4g 20440 MiB / 1244.12 / 76.008

表 2022-03-17 ResNet50 graph dlperf 运行结果

a33a825
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1481.12
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23100 MiB / 399.19
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23786 MiB / 1409.02
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23802 MiB / 395.39

表 2022-03-17 ResNet50 graph dlperf 运行结果

a33a825
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24104 MiB / 10461.52
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24248 MiB / 3114.88
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24876 MiB / 10315.6
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24916 MiB / 3056.88

表 2022-03-17 ResNet50 graph train 50E运行结果

a33a825
Memory/Throughput/TOP1
ResNet50_graph_train_gpudecode_FP16_b192_1n8g 11456 MiB / 10499.68 / 76.16199999999999

表 2022-03-17 ResNet50 graph train 50E运行结果

a33a825
Memory/Throughput/TOP1
ResNet50_graph_train_gpudecode_FP16_b192_1n4g 10754 MiB / 5215.8 / 76.204

表 2022-03-17 ResNet50 lazy dlperf 运行结果

a33a825
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23898 MiB / 1400.434
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23120 MiB / 1412.816
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23458 MiB / 382.435
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22700 MiB / 385.243

表 2022-03-17 ResNet50 lazy dlperf 运行结果

a33a825
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24490 MiB / 10445.925
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23720 MiB / 10988.424
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 23998 MiB / 2955.601
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23332 MiB / 3039.777

表 2022-03-18 WDL ddp dlperf 运行结果

4779c43
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5390 MiB / 301.6287130191922
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1800 MiB / 3.9399805702269077
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3516 MiB / 117.31615839898586

表 2022-03-18 WDL ddp dlperf 运行结果

4779c43
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2664 MiB / 41.66140232980251
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2214 MiB / 6.555784400552511
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2434 MiB / 21.011437561362982

表 2022-03-18 WDL graph dlperf 运行结果

4779c43
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4776 MiB / 182.61011029034853
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1506 MiB / 1.4958088546991348
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3136 MiB / 99.996641933918

表 2022-03-18 WDL graph dlperf 运行结果

4779c43
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2853 MiB / 21.99826715514064
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2135 MiB / 2.712374571710825
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2523 MiB / 12.237303487956524

表 2022-03-18 ResNet50 ddp dlperf 运行结果

a33a825
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 349.29

表 2022-03-18 ResNet50 ddp dlperf 运行结果

a33a825
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2512.24

表 2022-03-18 ResNet50 graph dlperf 运行结果

a33a825
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1459.65
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23172 MiB / 396.76
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23778 MiB / 1409.51
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23824 MiB / 395.13

表 2022-03-18 ResNet50 graph dlperf 运行结果

a33a825
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24104 MiB / 10358.4
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24154 MiB / 3105.76
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24902 MiB / 10737.2
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24988 MiB / 3015.12

表 2022-03-18 ResNet50 lazy dlperf 运行结果

a33a825
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23884 MiB / 1412.673
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1437.51
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23458 MiB / 380.015
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22700 MiB / 383.427

表 2022-03-18 ResNet50 lazy dlperf 运行结果

a33a825
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24490 MiB / 10634.348
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23722 MiB / 10971.626
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24050 MiB / 2967.226
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23336 MiB / 3035.054

表 2022-03-18 WDL ddp dlperf 运行结果

a33a825
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5390 MiB / 300.943211235106
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1800 MiB / 4.123374424874783
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3516 MiB / 118.2298629656434

表 2022-03-18 WDL ddp dlperf 运行结果

a33a825
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2664 MiB / 41.993391916155815
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2214 MiB / 6.476434085518122
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2434 MiB / 21.489576052874327

表 2022-03-18 WDL graph dlperf 运行结果

a33a825
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4776 MiB / 173.98804875835776
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1506 MiB / 1.5962225198745728
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3136 MiB / 98.26211120560765

表 2022-03-18 WDL graph dlperf 运行结果

a33a825
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2917 MiB / 21.86127570271492
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2135 MiB / 2.589229479432106
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2491 MiB / 12.066291097551584

表 2022-03-18 ResNet50 ddp dlperf 运行结果

d73fb1bad
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 357.21

表 2022-03-18 ResNet50 ddp dlperf 运行结果

d73fb1bad
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2512.56

表 2022-03-18 ResNet50 graph dlperf 运行结果

d73fb1bad
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1485.51
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23100 MiB / 405.78
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23806 MiB / 1425.29
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23866 MiB / 403.14

表 2022-03-18 ResNet50 graph dlperf 运行结果

d73fb1bad
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24124 MiB / 10535.84
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24204 MiB / 3104.56
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24840 MiB / 10304.48
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24998 MiB / 3028.08

表 2022-03-18 ResNet50 lazy dlperf 运行结果

d73fb1bad
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23904 MiB / 1437.011
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1453.927
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23420 MiB / 391.823
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22772 MiB / 391.443

表 2022-03-18 ResNet50 lazy dlperf 运行结果

d73fb1bad
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24514 MiB / 10600.176
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23720 MiB / 10877.477
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24036 MiB / 2959.605
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23280 MiB / 3040.553

表 2022-03-18 WDL ddp dlperf 运行结果

d73fb1bad
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5390 MiB / 303.7406092695892
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1800 MiB / 4.040969416499138
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3516 MiB / 114.95991970971227

表 2022-03-18 WDL ddp dlperf 运行结果

d73fb1bad
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2664 MiB / 40.74525924026966
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2214 MiB / 6.44284587725997
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2434 MiB / 21.36223767325282

表 2022-03-18 WDL graph dlperf 运行结果

d73fb1bad
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4776 MiB / 182.2251804023981
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1506 MiB / 1.5180705934762955
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 2880 MiB / 99.5955205745995

表 2022-03-18 WDL graph dlperf 运行结果

d73fb1bad
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2917 MiB / 21.835391085594893
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2135 MiB / 2.663074776530266
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2491 MiB / 12.09672874212265

表 2022-03-19 ResNet50 ddp dlperf 运行结果

2ccb633
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 348.7

表 2022-03-19 ResNet50 ddp dlperf 运行结果

2ccb633
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2552.4

表 2022-03-19 ResNet50 graph dlperf 运行结果

2ccb633
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1474.63
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23120 MiB / 401.0
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23772 MiB / 1405.82
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23862 MiB / 398.57

表 2022-03-19 ResNet50 graph dlperf 运行结果

2ccb633
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24124 MiB / 10580.8
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24210 MiB / 3114.48
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24850 MiB / 10794.88
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24946 MiB / 3072.56

表 2022-03-19 ResNet50 lazy dlperf 运行结果

2ccb633
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23892 MiB / 1421.643
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1419.515
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23444 MiB / 388.21
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22752 MiB / 389.547

表 2022-03-19 ResNet50 lazy dlperf 运行结果

2ccb633
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24500 MiB / 10477.691
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23722 MiB / 10977.943
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24010 MiB / 2986.072
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23308 MiB / 3038.208

表 2022-03-19 WDL ddp dlperf 运行结果

2ccb633
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5392 MiB / 305.69454441964626
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1802 MiB / 4.027995683252811
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3518 MiB / 121.41548044607043

表 2022-03-19 WDL ddp dlperf 运行结果

2ccb633
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2666 MiB / 41.677868600934744
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2216 MiB / 6.505902346223593
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2436 MiB / 21.043034818023443

表 2022-03-19 WDL graph dlperf 运行结果

2ccb633
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4778 MiB / 177.21335548907518
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1508 MiB / 1.5280469506978989
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3138 MiB / 98.67746280878782

表 2022-03-19 WDL graph dlperf 运行结果

2ccb633
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2919 MiB / 21.69344048574567
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2137 MiB / 2.6861617639660835
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2525 MiB / 12.088215507566929

表 2022-03-20 ResNet50 ddp dlperf 运行结果

f67aafb
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 345.04

表 2022-03-20 ResNet50 ddp dlperf 运行结果

f67aafb
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2533.84

表 2022-03-20 ResNet50 graph dlperf 运行结果

f67aafb
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1474.73
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23158 MiB / 394.08
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23768 MiB / 1414.21
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23874 MiB / 394.55

表 2022-03-20 ResNet50 graph dlperf 运行结果

f67aafb
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24104 MiB / 10588.72
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24156 MiB / 3118.48
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24882 MiB / 10808.56
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24902 MiB / 3063.12

表 2022-03-20 ResNet50 lazy dlperf 运行结果

f67aafb
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23908 MiB / 1419.512
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1429.564
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23436 MiB / 382.741
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22720 MiB / 383.944

表 2022-03-20 ResNet50 lazy dlperf 运行结果

f67aafb
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24492 MiB / 10562.453
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23724 MiB / 10870.208
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 23984 MiB / 2990.565
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23282 MiB / 3035.012

表 2022-03-20 WDL ddp dlperf 运行结果

f67aafb
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5392 MiB / 301.312696877867
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1802 MiB / 4.1110010370612144
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3518 MiB / 118.7471566349268

表 2022-03-20 WDL ddp dlperf 运行结果

f67aafb
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2666 MiB / 41.40678247064352
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2216 MiB / 6.623754560947418
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2436 MiB / 21.615786019712687

表 2022-03-20 WDL graph dlperf 运行结果

f67aafb
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4778 MiB / 177.59979778155684
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1510 MiB / 1.5120096988976002
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3074 MiB / 99.30825601890683

表 2022-03-20 WDL graph dlperf 运行结果

f67aafb
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2919 MiB / 21.003404833376408
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2137 MiB / 2.807922385632992
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2493 MiB / 12.129488002508879

表 2022-03-21 ResNet50 ddp dlperf 运行结果

b56df22
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 345.86

表 2022-03-21 ResNet50 graph dlperf 运行结果

b56df22
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1455.79
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23150 MiB / 396.14
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23776 MiB / 1426.23
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23816 MiB / 395.4

表 2022-03-21 ResNet50 lazy dlperf 运行结果

b56df22
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23884 MiB / 1405.803
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1424.803
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23482 MiB / 380.028
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22750 MiB / 382.997

表 2022-03-21 ResNet50 ddp dlperf 运行结果

f67aafb
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2546.8

表 2022-03-21 ResNet50 ddp train 50E运行结果

f67aafb
Memory/Throughput/TOP1
ResNet50_ddp_train_cpudecode_FP32_b192_1n4g 20492 MiB / 1297.04 / 76.152

表 2022-03-21 ResNet50 graph dlperf 运行结果

f67aafb
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24084 MiB / 10541.6
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24270 MiB / 3117.28
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24872 MiB / 10435.68
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24892 MiB / 3064.0

表 2022-03-21 ResNet50 graph train 50E运行结果

f67aafb
Memory/Throughput/TOP1
ResNet50_graph_train_gpudecode_FP16_b192_1n4g 10837 MiB / 5273.92 / 76.068

表 2022-03-21 ResNet50 lazy dlperf 运行结果

f67aafb
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24518 MiB / 10701.558
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23722 MiB / 10800.888
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24052 MiB / 2981.835
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23316 MiB / 3035.503

表 2022-03-21 WDL ddp dlperf 运行结果

f67aafb
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5392 MiB / 300.5122368372977
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1802 MiB / 3.9481681175529957
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3518 MiB / 113.28609575331211

表 2022-03-21 WDL ddp dlperf 运行结果

f67aafb
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2666 MiB / 41.93920290470123
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2216 MiB / 6.463845793157816
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2436 MiB / 21.160355649888515

表 2022-03-21 WDL graph dlperf 运行结果

f67aafb
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4778 MiB / 181.9131820164621
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1508 MiB / 1.540865670889616
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3138 MiB / 99.09145984053612

表 2022-03-21 WDL graph dlperf 运行结果

f67aafb
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2919 MiB / 22.035127319395542
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2137 MiB / 2.7336887568235397
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2525 MiB / 11.96633917838335

表 2022-03-22 ResNet50 ddp dlperf 运行结果

b56df22
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 355.05

表 2022-03-22 ResNet50 ddp dlperf 运行结果

b56df22
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2549.92

表 2022-03-22 ResNet50 graph dlperf 运行结果

b56df22
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1492.63
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23164 MiB / 408.44
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23776 MiB / 1436.37
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23820 MiB / 404.67

表 2022-03-22 ResNet50 graph dlperf 运行结果

b56df22
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24084 MiB / 10707.52
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24252 MiB / 3120.48
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24860 MiB / 10678.08
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24954 MiB / 3036.08

表 2022-03-22 ResNet50 lazy dlperf 运行结果

b56df22
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23916 MiB / 1436.964
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1424.949
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23462 MiB / 390.935
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22732 MiB / 393.343

表 2022-03-22 WDL ddp dlperf 运行结果

b56df22
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5392 MiB / 304.5288976803422
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1802 MiB / 4.084642864763737
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3518 MiB / 123.02796300500631

表 2022-03-22 WDL ddp dlperf 运行结果

b56df22
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2666 MiB / 41.25091652572155
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2216 MiB / 6.599991865456104
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2436 MiB / 21.357248045504093

表 2022-03-22 WDL graph dlperf 运行结果

b56df22
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4650 MiB / 170.595089122653
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1508 MiB / 1.4976569898426533
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3138 MiB / 97.20078879967332

表 2022-03-22 WDL graph dlperf 运行结果

b56df22
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2919 MiB / 21.93789129704237
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2137 MiB / 2.7141707874834538
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2493 MiB / 12.030795086175203

表 2022-03-23 ResNet50 ddp dlperf 运行结果

0d77085
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 348.07

表 2022-03-23 ResNet50 ddp dlperf 运行结果

0d77085
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2549.92

表 2022-03-23 ResNet50 graph dlperf 运行结果

0d77085
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1459.13
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23104 MiB / 398.3
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23766 MiB / 1400.39
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23804 MiB / 395.2

表 2022-03-23 ResNet50 graph dlperf 运行结果

0d77085
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24104 MiB / 10596.96
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24222 MiB / 3116.56
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24858 MiB / 10481.84
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24900 MiB / 3075.92

表 2022-03-23 ResNet50 lazy dlperf 运行结果

0d77085
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23884 MiB / 1404.764
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1426.094
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23414 MiB / 382.749
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22728 MiB / 384.945

表 2022-03-23 ResNet50 lazy dlperf 运行结果

0d77085
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24500 MiB / 10701.101
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23722 MiB / 10974.394
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24036 MiB / 2979.667
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23282 MiB / 3037.298

表 2022-03-23 WDL ddp dlperf 运行结果

0d77085
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5392 MiB / 305.03391606733203
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1802 MiB / 4.0297754518687725
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3518 MiB / 119.49975615739822

表 2022-03-23 WDL ddp dlperf 运行结果

0d77085
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2666 MiB / 40.79219709709287
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2216 MiB / 6.423699665814638
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2436 MiB / 21.563912618905306

表 2022-03-23 WDL graph dlperf 运行结果

0d77085
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4778 MiB / 173.41057139635086
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1508 MiB / 1.5375678576529026
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3138 MiB / 92.60580210760236

表 2022-03-23 WDL graph dlperf 运行结果

0d77085
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2855 MiB / 21.326160494238138
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2137 MiB / 2.711281441152096
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2525 MiB / 11.923645831644535

表 2022-03-27 ResNet50 ddp dlperf 运行结果

d873356b6
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 349.67

表 2022-03-27 ResNet50 ddp dlperf 运行结果

d873356b6
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2544.4

表 2022-03-27 ResNet50 graph dlperf 运行结果

d873356b6
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1482.71
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23150 MiB / 402.6
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23808 MiB / 1435.77
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23810 MiB / 396.87

表 2022-03-27 ResNet50 graph dlperf 运行结果

d873356b6
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24104 MiB / 10592.72
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24228 MiB / 3116.24
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24862 MiB / 10587.6
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24956 MiB / 3062.16

表 2022-03-27 ResNet50 lazy dlperf 运行结果

d873356b6
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23892 MiB / 1415.563
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1454.642
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23424 MiB / 393.55
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22748 MiB / 392.915

表 2022-03-27 ResNet50 lazy dlperf 运行结果

d873356b6
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24510 MiB / 10655.63
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23738 MiB / 10937.215
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24022 MiB / 2988.711
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23300 MiB / 3040.59

表 2022-03-28 ResNet50 ddp dlperf 运行结果

5a66b1b83
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 348.72

表 2022-03-28 ResNet50 ddp dlperf 运行结果

5a66b1b83
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2535.28

表 2022-03-28 ResNet50 graph dlperf 运行结果

5a66b1b83
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1460.28
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23102 MiB / 400.36
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23782 MiB / 1425.23
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23854 MiB / 398.51

表 2022-03-28 ResNet50 graph dlperf 运行结果

5a66b1b83
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24104 MiB / 10556.4
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24180 MiB / 3117.28
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24878 MiB / 10398.56
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24918 MiB / 3079.44

表 2022-03-28 ResNet50 lazy dlperf 运行结果

5a66b1b83
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23902 MiB / 1414.275
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23132 MiB / 1443.223
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23470 MiB / 388.75
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22748 MiB / 391.681

表 2022-03-28 ResNet50 lazy dlperf 运行结果

5a66b1b83
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24546 MiB / 10771.664
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23722 MiB / 11006.065
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24030 MiB / 2957.329
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23260 MiB / 3035.599

表 2022-03-29 ResNet50 ddp dlperf 运行结果

0f45edd
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2539.04

表 2022-03-29 ResNet50 graph dlperf 运行结果

0f45edd
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24084 MiB / 10746.96
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24208 MiB / 3121.04
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24880 MiB / 10719.68
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24938 MiB / 3071.12

表 2022-03-29 ResNet50 lazy dlperf 运行结果

0f45edd
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24492 MiB / 10711.979
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23724 MiB / 10909.141
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24014 MiB / 2976.101
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23308 MiB / 3036.257

表 2022-03-29 WDL ddp dlperf 运行结果

0f45edd
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2666 MiB / 41.54586097598076
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2216 MiB / 6.562971413135529
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2436 MiB / 21.19541274383664

表 2022-03-29 WDL graph dlperf 运行结果

0f45edd
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2919 MiB / 21.363861203193665
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2137 MiB / 2.623606711626053
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2525 MiB / 12.086092356592417

表 2022-03-29 ResNet50 ddp dlperf 运行结果

24543a5
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 342.38

表 2022-03-29 ResNet50 graph dlperf 运行结果

24543a5
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1474.78
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23154 MiB / 397.22
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23780 MiB / 1424.81
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23824 MiB / 395.02

表 2022-03-29 ResNet50 lazy dlperf 运行结果

24543a5
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23902 MiB / 1408.18
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1426.241
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23474 MiB / 379.223
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22760 MiB / 382.521

表 2022-03-29 WDL ddp dlperf 运行结果

24543a5
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5392 MiB / 302.6602213829756
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1802 MiB / 4.083907350897789
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3518 MiB / 126.41270082071424

表 2022-03-29 WDL graph dlperf 运行结果

24543a5
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4778 MiB / 161.382267344743
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1510 MiB / 1.4903183579444885
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3138 MiB / 93.55231199041009

表 2022-03-30 ResNet50 ddp dlperf 运行结果

0f45edd
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 349.38

表 2022-03-30 ResNet50 graph dlperf 运行结果

0f45edd
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1480.79
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23174 MiB / 399.54
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23776 MiB / 1419.69
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23908 MiB / 396.48

表 2022-03-30 ResNet50 lazy dlperf 运行结果

0f45edd
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23896 MiB / 1417.703
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1411.982
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23434 MiB / 384.205
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22756 MiB / 386.821

表 2022-03-30 ResNet50 ddp dlperf 运行结果

c7fc63421
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 346.11

表 2022-03-30 ResNet50 ddp dlperf 运行结果

c7fc63421
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2537.36

表 2022-03-30 ResNet50 graph dlperf 运行结果

c7fc63421
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1470.86
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23096 MiB / 396.56
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23782 MiB / 1471.66
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23872 MiB / 393.9

表 2022-03-30 ResNet50 graph dlperf 运行结果

c7fc63421
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24104 MiB / 10687.12
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24174 MiB / 3107.44
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24866 MiB / 10566.64
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24978 MiB / 3050.88

表 2022-03-30 ResNet50 lazy dlperf 运行结果

c7fc63421
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23884 MiB / 1399.12
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1425.162
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23478 MiB / 380.248
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22700 MiB / 382.893

表 2022-03-30 ResNet50 lazy dlperf 运行结果

c7fc63421
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24488 MiB / 10595.254
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23722 MiB / 10895.295
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24022 MiB / 2969.733
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23276 MiB / 3033.269

表 2022-03-31 ResNet50 ddp dlperf 运行结果

6045c4f1e
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27370 MiB / 2534.56

表 2022-03-31 ResNet50 graph dlperf 运行结果

6045c4f1e
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24084 MiB / 10678.24
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24240 MiB / 3109.52
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24870 MiB / 10069.28
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24936 MiB / 3065.04

表 2022-03-31 ResNet50 lazy dlperf 运行结果

6045c4f1e
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n8g 24504 MiB / 10729.363
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n8g 23720 MiB / 10872.467
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n8g 24006 MiB / 2992.309
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n8g 23318 MiB / 3037.84

表 2022-03-31 WDL ddp dlperf 运行结果

b80a825
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5392 MiB / 304.59448911994696
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1802 MiB / 4.143860567361116
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3518 MiB / 118.1583112590015

表 2022-03-31 WDL ddp dlperf 运行结果

b80a825
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2666 MiB / 41.980448897928
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2216 MiB / 6.521544016897678
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2436 MiB / 21.13132306933403

表 2022-03-31 WDL graph dlperf 运行结果

b80a825
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4778 MiB / 182.08862436190248
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1508 MiB / 1.4595725163817406
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3138 MiB / 99.39803534001112

表 2022-03-31 WDL graph dlperf 运行结果

b80a825
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2855 MiB / 21.77755380794406
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2137 MiB / 2.6841637641191483
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2493 MiB / 11.932186890393496

表 2022-03-31 ResNet50 ddp dlperf 运行结果

cdae64a
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26938 MiB / 350.67

表 2022-03-31 ResNet50 graph dlperf 运行结果

cdae64a
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23006 MiB / 1498.63
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23150 MiB / 398.72
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23790 MiB / 1467.91
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23846 MiB / 395.4

表 2022-03-31 ResNet50 lazy dlperf 运行结果

cdae64a
Memory/Throughput
ResNet50_lazy_dlperf_gpudecode_FP16_b512_1n1g 23892 MiB / 1419.381
ResNet50_lazy_dlperf_cpudecode_FP16_b512_1n1g 23118 MiB / 1425.143
ResNet50_lazy_dlperf_gpudecode_FP32_b256_1n1g 23488 MiB / 383.298
ResNet50_lazy_dlperf_cpudecode_FP32_b256_1n1g 22720 MiB / 387.144

表 2022-03-31 WDL ddp dlperf 运行结果

cdae64a
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5392 MiB / 303.5307957492769
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1802 MiB / 4.083432219922543
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3518 MiB / 121.55172082409263

表 2022-03-31 WDL graph dlperf 运行结果

cdae64a
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4778 MiB / 161.5074171088636
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1508 MiB / 1.4948842972517014
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3138 MiB / 92.85725738853216