Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

202206_OneFlow_All_Models_Test_Report_On_master #26

Open
github-actions bot opened this issue Jun 7, 2022 · 22 comments
Open

202206_OneFlow_All_Models_Test_Report_On_master #26

github-actions bot opened this issue Jun 7, 2022 · 22 comments

Comments

@github-actions
Copy link

github-actions bot commented Jun 7, 2022

202206_OneFlow_All_Models_Test_Report_On_master

@github-actions
Copy link
Author

github-actions bot commented Jun 7, 2022

表 2022-06-07 ResNet50 ddp dlperf 运行结果

83ed0ba5
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27342 MiB / 2354.88

表 2022-06-07 ResNet50 graph dlperf 运行结果

83ed0ba5
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24058 MiB / 10842.32
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24166 MiB / 3030.56
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24864 MiB / 10615.92
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24896 MiB / 2981.84

@github-actions
Copy link
Author

github-actions bot commented Jun 8, 2022

表 2022-06-08 ResNet50 ddp dlperf 运行结果

7feedc44
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26942 MiB / 317.46

表 2022-06-08 ResNet50 ddp dlperf 运行结果

7feedc44
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27342 MiB / 2347.28

表 2022-06-08 ResNet50 graph dlperf 运行结果

7feedc44
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22914 MiB / 1517.36
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 22996 MiB / 387.64
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23700 MiB / 1507.07
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23730 MiB / 385.7

表 2022-06-08 ResNet50 graph dlperf 运行结果

7feedc44
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24076 MiB / 10872.88
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24166 MiB / 3027.68
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24872 MiB / 10613.04
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24902 MiB / 2944.8

表 2022-06-08 WDL ddp dlperf 运行结果

7feedc44
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5626 MiB / 300.50578474998474
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1826 MiB / 4.660750225186348
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3812 MiB / 120.5880561992526

表 2022-06-08 WDL ddp dlperf 运行结果

7feedc44
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2700 MiB / 41.11755032837391
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2226 MiB / 8.025342307984829
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2438 MiB / 22.336046755313873

表 2022-06-08 WDL graph dlperf 运行结果

7feedc44
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4754 MiB / 163.47150307148695
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 2858 MiB / 94.49771839380264

表 2022-06-08 WDL graph dlperf 运行结果

7feedc44
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 3995 MiB / 23.893178574740887
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2849 MiB / 4.251195523887873
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 3449 MiB / 13.946942273527384

@github-actions
Copy link
Author

github-actions bot commented Jun 9, 2022

表 2022-06-09 ResNet50 ddp dlperf 运行结果

0fd888c
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26942 MiB / 319.23

表 2022-06-09 ResNet50 ddp dlperf 运行结果

0fd888c
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27342 MiB / 2344.4

表 2022-06-09 ResNet50 graph dlperf 运行结果

0fd888c
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22914 MiB / 1519.63
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23044 MiB / 388.13
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23696 MiB / 1509.31
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23798 MiB / 384.91

表 2022-06-09 ResNet50 graph dlperf 运行结果

0fd888c
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24076 MiB / 10779.12
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24144 MiB / 3031.36
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24850 MiB / 10718.96
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24916 MiB / 2977.84

表 2022-06-09 WDL ddp dlperf 运行结果

0fd888c
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5628 MiB / 300.58381901681423
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1828 MiB / 4.726193316280842
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3814 MiB / 121.58754910156131

表 2022-06-09 WDL ddp dlperf 运行结果

0fd888c
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2702 MiB / 41.41168262809515
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2228 MiB / 8.079352527856827
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2440 MiB / 22.054394774138927

表 2022-06-09 WDL graph dlperf 运行结果

0fd888c
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4756 MiB / 161.28383708372712
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 2860 MiB / 97.31046555563807

表 2022-06-09 WDL graph dlperf 运行结果

0fd888c
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 3997 MiB / 23.6545618660748
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2851 MiB / 4.23183373734355
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 3419 MiB / 13.879558328539133

@github-actions
Copy link
Author

表 2022-06-10 ResNet50 ddp dlperf 运行结果

469f72d8
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26942 MiB / 318.58

表 2022-06-10 ResNet50 ddp dlperf 运行结果

469f72d8
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27342 MiB / 2346.8

表 2022-06-10 ResNet50 graph dlperf 运行结果

469f72d8
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22914 MiB / 1509.85
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 22998 MiB / 388.33
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23696 MiB / 1500.94
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23760 MiB / 384.8

表 2022-06-10 ResNet50 graph dlperf 运行结果

469f72d8
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24076 MiB / 10764.72
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24186 MiB / 3036.0
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24840 MiB / 10618.16
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24858 MiB / 2974.88

表 2022-06-10 WDL ddp dlperf 运行结果

d98132ee
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5628 MiB / 301.41041019931436
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1828 MiB / 4.745050612837076
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3814 MiB / 117.93160200119019

表 2022-06-10 WDL ddp dlperf 运行结果

d98132ee
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2702 MiB / 41.15106539428234
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2228 MiB / 7.931446939706802
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2440 MiB / 21.25645286217332

表 2022-06-10 WDL graph dlperf 运行结果

d98132ee
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4756 MiB / 164.9041360579431
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 2860 MiB / 93.88274237513542

表 2022-06-10 WDL graph dlperf 运行结果

d98132ee
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 3997 MiB / 23.657179471105337
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2851 MiB / 4.241833500564098
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 3419 MiB / 13.86045167595148

@github-actions
Copy link
Author

表 2022-06-11 ResNet50 ddp dlperf 运行结果

5686581c
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26942 MiB / 317.66

表 2022-06-11 ResNet50 ddp dlperf 运行结果

5686581c
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27342 MiB / 2350.56

表 2022-06-11 ResNet50 graph dlperf 运行结果

5686581c
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22914 MiB / 1506.34
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23044 MiB / 386.67
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23696 MiB / 1490.3
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23734 MiB / 385.95

表 2022-06-11 ResNet50 graph dlperf 运行结果

5686581c
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24058 MiB / 10656.48
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24186 MiB / 3034.08
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24852 MiB / 10932.08
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24922 MiB / 2966.56

表 2022-06-11 WDL ddp dlperf 运行结果

5686581c
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5628 MiB / 300.69503996148705
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1828 MiB / 4.451040334999561
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3814 MiB / 118.88804276660085

表 2022-06-11 WDL ddp dlperf 运行结果

5686581c
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2702 MiB / 41.42889918759465
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2228 MiB / 8.100533060729504
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2440 MiB / 22.254176136106253

表 2022-06-11 WDL graph dlperf 运行结果

5686581c
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4756 MiB / 166.46078354120255
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3116 MiB / 90.89364966377616

表 2022-06-11 WDL graph dlperf 运行结果

5686581c
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 4061 MiB / 23.86993148177862
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2851 MiB / 4.245437238365412
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 3419 MiB / 13.979482892900705

@github-actions
Copy link
Author

表 2022-06-12 ResNet50 ddp dlperf 运行结果

15eef189
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26942 MiB / 318.79

表 2022-06-12 ResNet50 ddp dlperf 运行结果

15eef189
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27342 MiB / 2342.8

表 2022-06-12 ResNet50 graph dlperf 运行结果

15eef189
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22914 MiB / 1515.21
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23042 MiB / 388.51
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23702 MiB / 1501.69
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23762 MiB / 384.32

表 2022-06-12 ResNet50 graph dlperf 运行结果

15eef189
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24054 MiB / 10702.4
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24158 MiB / 3030.08
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24830 MiB / 10925.12
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24876 MiB / 2986.24

表 2022-06-12 WDL ddp dlperf 运行结果

15eef189
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2702 MiB / 41.705647833645344
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2228 MiB / 7.957782126963139
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2440 MiB / 22.0493346452713

表 2022-06-12 WDL graph dlperf 运行结果

15eef189
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 4061 MiB / 23.769599556922913
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2851 MiB / 4.197185583412647
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 3419 MiB / 13.77873283997178

@github-actions
Copy link
Author

表 2022-06-13 ResNet50 ddp dlperf 运行结果

15eef189
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26942 MiB / 318.84

表 2022-06-13 ResNet50 ddp dlperf 运行结果

15eef189
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27342 MiB / 2352.64

表 2022-06-13 ResNet50 graph dlperf 运行结果

15eef189
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22914 MiB / 1520.57
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23032 MiB / 388.32
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23694 MiB / 1499.5
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23704 MiB / 386.47

表 2022-06-13 ResNet50 graph dlperf 运行结果

15eef189
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24076 MiB / 10804.0
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24124 MiB / 3036.4
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24884 MiB / 10779.04
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24910 MiB / 2956.24

表 2022-06-13 WDL ddp dlperf 运行结果

15eef189
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5628 MiB / 301.2053214684129
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1828 MiB / 4.866649553179741
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3814 MiB / 119.1432052589953

表 2022-06-13 WDL ddp dlperf 运行结果

15eef189
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2702 MiB / 41.27906038239598
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2228 MiB / 8.01334785297513
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2440 MiB / 21.600046645849943

表 2022-06-13 WDL graph dlperf 运行结果

15eef189
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4246 MiB / 166.49398171156645
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3116 MiB / 91.69520938023925

表 2022-06-13 WDL graph dlperf 运行结果

15eef189
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 4061 MiB / 23.584800109267235
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2851 MiB / 4.245323937386274
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 3451 MiB / 13.970636077225208

表 2022-06-13 ResNet50 ddp train 50E运行结果

39a2d45
Memory/Throughput/TOP1
ResNet50_ddp_train_cpudecode_FP32_b192_1n4g 20882 MiB / 1196.56 / 76.112

表 2022-06-13 ResNet50 graph train 50E运行结果

39a2d45
Memory/Throughput/TOP1
ResNet50_graph_train_gpudecode_FP16_b192_1n4g 10684 MiB / 5292.4 / 76.036

@github-actions
Copy link
Author

表 2022-06-14 ResNet50 ddp dlperf 运行结果

e8547b45
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26942 MiB / 319.58

表 2022-06-14 ResNet50 ddp dlperf 运行结果

e8547b45
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27342 MiB / 2343.76

表 2022-06-14 ResNet50 graph dlperf 运行结果

e8547b45
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22918 MiB / 1524.02
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23048 MiB / 389.89
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23682 MiB / 1505.22
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23730 MiB / 388.23

表 2022-06-14 ResNet50 graph dlperf 运行结果

e8547b45
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24060 MiB / 10862.32
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24176 MiB / 3027.12
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24824 MiB / 11099.68
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24884 MiB / 2962.48

表 2022-06-14 WDL ddp dlperf 运行结果

e8547b45
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5628 MiB / 303.10341189056635
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1828 MiB / 4.717332907021046
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3814 MiB / 120.23116274923086

表 2022-06-14 WDL ddp dlperf 运行结果

e8547b45
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2702 MiB / 40.688297133892775
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2228 MiB / 7.878821682184935
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2440 MiB / 21.94750927388668

表 2022-06-14 WDL graph dlperf 运行结果

e8547b45
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4756 MiB / 163.1213648803532
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3116 MiB / 93.52434170246124

表 2022-06-14 WDL graph dlperf 运行结果

e8547b45
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 4059 MiB / 23.681406520307064
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2849 MiB / 4.228797543793917
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 3417 MiB / 13.87756635248661

@github-actions
Copy link
Author

表 2022-06-15 ResNet50 ddp dlperf 运行结果

041f7871
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26942 MiB / 319.55

表 2022-06-15 ResNet50 graph dlperf 运行结果

041f7871
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22918 MiB / 1524.92
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23020 MiB / 391.21
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23696 MiB / 1508.35
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23742 MiB / 389.02

表 2022-06-15 WDL ddp dlperf 运行结果

041f7871
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5628 MiB / 301.20594861730933
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1828 MiB / 4.39649211242795
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3814 MiB / 113.75622241199017

表 2022-06-15 WDL graph dlperf 运行结果

041f7871
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4246 MiB / 177.84363690018654
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3116 MiB / 99.39724956080317

@github-actions
Copy link
Author

表 2022-06-16 WDL ddp dlperf 运行结果

4af9b7a4
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5628 MiB / 300.91312374547124
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1828 MiB / 4.335263378918171
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3814 MiB / 112.59728172793984

表 2022-06-16 WDL graph dlperf 运行结果

4af9b7a4
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4756 MiB / 183.4780812598765
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3116 MiB / 99.92728035897017

@github-actions
Copy link
Author

表 2022-06-17 ResNet50 ddp dlperf 运行结果

f6c3cb6
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26942 MiB / 318.6

表 2022-06-17 ResNet50 ddp dlperf 运行结果

f6c3cb6
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27342 MiB / 2351.2

表 2022-06-17 ResNet50 graph dlperf 运行结果

f6c3cb6
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22918 MiB / 1510.79
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23068 MiB / 386.63
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23684 MiB / 1482.02
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23776 MiB / 383.62

表 2022-06-17 ResNet50 graph dlperf 运行结果

f6c3cb6
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24082 MiB / 10786.56
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24202 MiB / 3037.28
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24848 MiB / 10919.04
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24908 MiB / 2977.52

表 2022-06-17 WDL ddp dlperf 运行结果

f6c3cb6
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5628 MiB / 301.41143449395895
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1828 MiB / 4.629772141575813
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3814 MiB / 117.88295146077871

表 2022-06-17 WDL graph dlperf 运行结果

f6c3cb6
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4756 MiB / 163.60815526917577
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3116 MiB / 95.9946751408279

@github-actions
Copy link
Author

表 2022-06-18 ResNet50 ddp dlperf 运行结果

f7532fd
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26942 MiB / 318.33

表 2022-06-18 ResNet50 ddp dlperf 运行结果

f7532fd
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27342 MiB / 2360.72

表 2022-06-18 ResNet50 graph dlperf 运行结果

f7532fd
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22914 MiB / 1497.51
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23038 MiB / 387.12
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23698 MiB / 1480.54
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23756 MiB / 384.98

表 2022-06-18 ResNet50 graph dlperf 运行结果

f7532fd
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24060 MiB / 10952.4
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24112 MiB / 3037.68
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24824 MiB / 11087.92
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24936 MiB / 2955.36

表 2022-06-18 WDL ddp dlperf 运行结果

f7532fd
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5628 MiB / 303.4966679662466
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1828 MiB / 4.5234799683094025
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3814 MiB / 115.59121402725577

表 2022-06-18 WDL graph dlperf 运行结果

f7532fd
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5146 MiB / 172.67607191950083
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1642 MiB / 2.562586944550276
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g None MiB / 0

@github-actions
Copy link
Author

表 2022-06-19 ResNet50 ddp dlperf 运行结果

d7ef39f
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 26942 MiB / 318.6

表 2022-06-19 ResNet50 graph dlperf 运行结果

d7ef39f
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22912 MiB / 1505.49
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23038 MiB / 385.51
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23694 MiB / 1483.49
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23774 MiB / 385.39

表 2022-06-19 WDL ddp dlperf 运行结果

d7ef39f
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5628 MiB / 301.1190925599949
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1828 MiB / 4.654642124005477
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3814 MiB / 111.86599666600523

表 2022-06-19 WDL graph dlperf 运行结果

d7ef39f
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5146 MiB / 182.78035577500123
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1642 MiB / 2.6110918080012198
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g None MiB / 0

@github-actions
Copy link
Author

表 2022-06-22 ResNet50 ddp dlperf 运行结果

3cbf392
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27004 MiB / 318.81

表 2022-06-22 ResNet50 ddp dlperf 运行结果

3cbf392
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27404 MiB / 2349.2

表 2022-06-22 ResNet50 graph dlperf 运行结果

3cbf392
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22976 MiB / 1530.29
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23098 MiB / 396.17
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23740 MiB / 1516.74
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23824 MiB / 394.67

表 2022-06-22 ResNet50 graph dlperf 运行结果

3cbf392
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24124 MiB / 10911.92
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24226 MiB / 3045.6
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24892 MiB / 10604.88
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24950 MiB / 3011.12

表 2022-06-22 WDL ddp dlperf 运行结果

3cbf392
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5688 MiB / 301.321766089648
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1888 MiB / 4.4015613459050655
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3874 MiB / 119.71928597614169

表 2022-06-22 WDL ddp dlperf 运行结果

3cbf392
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2762 MiB / 41.31327151879668
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2288 MiB / 7.919158041477203
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2500 MiB / 21.97400389239192

表 2022-06-22 WDL graph dlperf 运行结果

3cbf392
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4822 MiB / 172.58456060290337
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1702 MiB / 2.6947810240089893
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g None MiB / 0

表 2022-06-22 WDL graph dlperf 运行结果

3cbf392
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 4119 MiB / 23.958338905125856
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g None MiB / 0

表 2022-06-22 ResNet50 ddp dlperf 运行结果

ca3cbdd
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27004 MiB / 318.93

表 2022-06-22 ResNet50 graph dlperf 运行结果

ca3cbdd
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22976 MiB / 1511.86
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23096 MiB / 392.38
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23760 MiB / 1490.37
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23820 MiB / 388.27

@github-actions
Copy link
Author

表 2022-06-23 ResNet50 ddp dlperf 运行结果

9a5e7506
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27004 MiB / 319.03

表 2022-06-23 ResNet50 graph dlperf 运行结果

9a5e7506
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22976 MiB / 1502.63
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23096 MiB / 389.81
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23750 MiB / 1480.35
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23828 MiB / 387.26

表 2022-06-23 WDL ddp dlperf 运行结果

9a5e7506
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5688 MiB / 302.4031555049587
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1888 MiB / 4.506519008951727
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3874 MiB / 115.69610099302372

表 2022-06-23 WDL graph dlperf 运行结果

9a5e7506
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4822 MiB / 173.6790693840012
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1702 MiB / 2.7235617639962584
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g None MiB / 0

@github-actions
Copy link
Author

表 2022-06-24 WDL ddp dlperf 运行结果

3b90b60
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2762 MiB / 79.49534908198984
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2288 MiB / 10.421653264027555
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2500 MiB / 42.39187942299759

表 2022-06-24 WDL graph dlperf 运行结果

3b90b60
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2909 MiB / 4.119705057004467
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 3509 MiB / 13.712993313965853

表 2022-06-24 ResNet50 ddp train 50E运行结果

a224698
Memory/Throughput/TOP1
ResNet50_ddp_train_cpudecode_FP32_b192_1n4g 21048 MiB / 1167.0 / 76.412

表 2022-06-24 ResNet50 graph train 50E运行结果

a224698
Memory/Throughput/TOP1
ResNet50_graph_train_gpudecode_FP16_b192_1n4g 10986 MiB / 5343.8 / 75.788

表 2022-06-24 ResNet50 ddp dlperf 运行结果

a4a20016
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27004 MiB / 319.31

表 2022-06-24 ResNet50 ddp dlperf 运行结果

a4a20016
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27404 MiB / 2352.88

表 2022-06-24 ResNet50 graph dlperf 运行结果

a4a20016
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22976 MiB / 1505.56
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23098 MiB / 389.11
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23740 MiB / 1486.33
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23832 MiB / 386.54

表 2022-06-24 ResNet50 graph dlperf 运行结果

a4a20016
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24122 MiB / 11022.8
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24226 MiB / 3056.4
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24922 MiB / 11093.28
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24918 MiB / 2908.08

表 2022-06-24 WDL ddp dlperf 运行结果

a4a20016
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5688 MiB / 301.41664287802996
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1888 MiB / 4.341844181995839
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3874 MiB / 115.98759194696322

表 2022-06-24 WDL graph dlperf 运行结果

a4a20016
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4822 MiB / 174.60023124102736
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1702 MiB / 2.6562046500039287
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g None MiB / 0

@github-actions
Copy link
Author

表 2022-06-25 ResNet50 ddp dlperf 运行结果

26fe902
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27004 MiB / 320.28

表 2022-06-25 ResNet50 ddp dlperf 运行结果

26fe902
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27404 MiB / 2373.44

表 2022-06-25 ResNet50 graph dlperf 运行结果

26fe902
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22976 MiB / 1523.86
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23098 MiB / 392.79
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23740 MiB / 1489.09
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23836 MiB / 391.54

表 2022-06-25 ResNet50 graph dlperf 运行结果

26fe902
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24078 MiB / 10991.44
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24182 MiB / 3060.8
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24848 MiB / 11075.92
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24890 MiB / 3023.28

表 2022-06-25 WDL ddp dlperf 运行结果

26fe902
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5688 MiB / 538.8195799551904
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1888 MiB / 7.327823620289564
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3874 MiB / 221.0114262998104

表 2022-06-25 WDL ddp dlperf 运行结果

26fe902
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2762 MiB / 80.382591897971
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2288 MiB / 10.492053756956011
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2500 MiB / 43.12843960901955

表 2022-06-25 WDL graph dlperf 运行结果

26fe902
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4822 MiB / 176.19572351127863
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1702 MiB / 2.714967180043459
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g None MiB / 0

表 2022-06-25 WDL graph dlperf 运行结果

26fe902
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 3489 MiB / 13.908154460950755

@github-actions
Copy link
Author

表 2022-06-26 ResNet50 ddp train 50E运行结果

3ea445a
Memory/Throughput/TOP1
ResNet50_ddp_train_cpudecode_FP32_b192_1n8g 21332 MiB / 2314.72 / 76.232

表 2022-06-26 ResNet50 graph train 50E运行结果

3ea445a
Memory/Throughput/TOP1
ResNet50_graph_train_gpudecode_FP16_b192_1n8g 11480 MiB / 10831.76 / 76.11000000000001

表 2022-06-26 ResNet50 ddp dlperf 运行结果

ca9fd64
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27004 MiB / 320.03

表 2022-06-26 ResNet50 ddp dlperf 运行结果

ca9fd64
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27404 MiB / 2384.08

表 2022-06-26 ResNet50 graph dlperf 运行结果

ca9fd64
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22976 MiB / 1513.4
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23098 MiB / 390.18
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23740 MiB / 1491.98
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23814 MiB / 387.0

表 2022-06-26 ResNet50 graph dlperf 运行结果

ca9fd64
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24080 MiB / 11082.24
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24178 MiB / 3046.64
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24856 MiB / 10675.92
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24926 MiB / 2980.64

表 2022-06-26 WDL ddp dlperf 运行结果

ca9fd64
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5690 MiB / 524.22232940502
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1890 MiB / 7.348110530991107
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3876 MiB / 231.46010174695402

表 2022-06-26 WDL ddp dlperf 运行结果

ca9fd64
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2764 MiB / 79.74042892595753
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2290 MiB / 10.365586945903488
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2502 MiB / 43.322030208073556

表 2022-06-26 WDL graph dlperf 运行结果

ca9fd64
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4824 MiB / 165.88101231097244
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1704 MiB / 2.688529178034514
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 2938 MiB / 96.69531977607403

表 2022-06-26 WDL graph dlperf 运行结果

ca9fd64
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2891 MiB / 4.0912869040621445
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g None MiB / 0

@github-actions
Copy link
Author

表 2022-06-27 ResNet50 ddp dlperf 运行结果

0c35fd17
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27004 MiB / 319.81

表 2022-06-27 ResNet50 ddp dlperf 运行结果

0c35fd17
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27404 MiB / 2383.36

表 2022-06-27 ResNet50 graph dlperf 运行结果

0c35fd17
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22980 MiB / 1518.56
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23104 MiB / 392.47
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23748 MiB / 1500.12
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23828 MiB / 389.5

表 2022-06-27 ResNet50 graph dlperf 运行结果

0c35fd17
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24078 MiB / 11084.24
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24178 MiB / 3055.44
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24850 MiB / 10437.6
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24922 MiB / 2974.56

表 2022-06-27 WDL ddp dlperf 运行结果

0c35fd17
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5690 MiB / 501.3365491509903
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1890 MiB / 7.059446800965816
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3876 MiB / 219.6264444469707

表 2022-06-27 WDL ddp dlperf 运行结果

0c35fd17
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2764 MiB / 79.54279957397375
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2290 MiB / 10.42214438796509
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2502 MiB / 43.17570986796636

表 2022-06-27 WDL graph dlperf 运行结果

0c35fd17
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4818 MiB / 168.30583986907732
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1702 MiB / 2.6971728180069476
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3178 MiB / 94.48766555893235

表 2022-06-27 WDL graph dlperf 运行结果

0c35fd17
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 4101 MiB / 24.156991750001907
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2891 MiB / 4.125056855962612
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 3491 MiB / 13.802654539002106

表 2022-06-27 ResNet50 ddp train 50E运行结果

a224698
Memory/Throughput/TOP1
ResNet50_ddp_train_cpudecode_FP32_b192_1n4g 21064 MiB / 1162.04 / 38.12

表 2022-06-27 ResNet50 graph train 50E运行结果

a224698
Memory/Throughput/TOP1
ResNet50_graph_train_gpudecode_FP16_b192_1n4g 10937 MiB / 5350.6 / 76.216

@github-actions
Copy link
Author

表 2022-06-28 ResNet50 ddp dlperf 运行结果

f52ec97
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27004 MiB / 320.09

表 2022-06-28 ResNet50 ddp dlperf 运行结果

f52ec97
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27404 MiB / 2383.2

表 2022-06-28 ResNet50 graph dlperf 运行结果

f52ec97
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22980 MiB / 1525.79
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23104 MiB / 391.82
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23766 MiB / 1496.31
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23820 MiB / 390.72

表 2022-06-28 ResNet50 graph dlperf 运行结果

f52ec97
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24078 MiB / 11024.16
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24134 MiB / 3038.64
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24866 MiB / 10647.28
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24906 MiB / 2976.08

表 2022-06-28 WDL ddp dlperf 运行结果

f52ec97
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5690 MiB / 524.0951528720325
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1890 MiB / 7.311824126983993
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3876 MiB / 222.68406562309247

表 2022-06-28 WDL ddp dlperf 运行结果

f52ec97
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2764 MiB / 80.10335256159306
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2290 MiB / 10.392899833619595
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2502 MiB / 42.6163682974875

表 2022-06-28 WDL graph dlperf 运行结果

f52ec97
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4818 MiB / 173.7956370320171
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1702 MiB / 2.6835626820102334
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3178 MiB / 94.90687304700259

表 2022-06-28 WDL graph dlperf 运行结果

f52ec97
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g None MiB / 0
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2891 MiB / 4.091257683932781
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g None MiB / 0

@github-actions
Copy link
Author

表 2022-06-29 ResNet50 ddp dlperf 运行结果

2aaa2a9
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27032 MiB / 318.66

表 2022-06-29 ResNet50 graph dlperf 运行结果

2aaa2a9
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23008 MiB / 1512.09
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23132 MiB / 388.25
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23774 MiB / 1500.54
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23852 MiB / 385.29

表 2022-06-29 WDL ddp dlperf 运行结果

2aaa2a9
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5718 MiB / 514.7828076556325
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1918 MiB / 7.2053748816251755
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3904 MiB / 228.02184193208814

表 2022-06-29 WDL graph dlperf 运行结果

2aaa2a9
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4336 MiB / 162.45167232304811
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1730 MiB / 2.5593802519142628
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 2950 MiB / 89.58061415702105

@github-actions
Copy link
Author

表 2022-06-30 ResNet50 ddp dlperf 运行结果

327a19fb
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27432 MiB / 2374.48

表 2022-06-30 ResNet50 graph dlperf 运行结果

327a19fb
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24106 MiB / 10796.48
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24210 MiB / 3045.6
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24864 MiB / 10986.16
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24938 MiB / 2961.28

表 2022-06-30 WDL ddp dlperf 运行结果

327a19fb
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2792 MiB / 77.76998557406478
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2318 MiB / 10.406715307035483
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2530 MiB / 42.920583332073875

表 2022-06-30 WDL graph dlperf 运行结果

327a19fb
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 4065 MiB / 24.096481892978773
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2919 MiB / 4.213190328911878
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 3487 MiB / 13.868803899968043

表 2022-06-30 ResNet50 ddp dlperf 运行结果

99b1eff7
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27032 MiB / 319.05

表 2022-06-30 ResNet50 graph dlperf 运行结果

99b1eff7
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 23008 MiB / 1513.19
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23132 MiB / 387.85
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23788 MiB / 1490.39
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23858 MiB / 388.46

表 2022-06-30 WDL ddp dlperf 运行结果

99b1eff7
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5718 MiB / 533.2870297139743
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1918 MiB / 7.023936112993397
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3904 MiB / 224.21730228001252

表 2022-06-30 WDL graph dlperf 运行结果

99b1eff7
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4336 MiB / 175.67936503503006
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1730 MiB / 2.6659722849726677
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 2950 MiB / 92.07955204101745

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

0 participants