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

202205_OneFlow_All_Models_Test_Report_On_master #23

Open
github-actions bot opened this issue May 2, 2022 · 20 comments
Open

202205_OneFlow_All_Models_Test_Report_On_master #23

github-actions bot opened this issue May 2, 2022 · 20 comments

Comments

@github-actions
Copy link

github-actions bot commented May 2, 2022

202205_OneFlow_All_Models_Test_Report_On_master

@github-actions
Copy link
Author

github-actions bot commented May 2, 2022

表 2022-05-02 ResNet50 ddp dlperf 运行结果

bc3af66a
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27060 MiB / 319.26

表 2022-05-02 ResNet50 ddp dlperf 运行结果

bc3af66a
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27460 MiB / 2350.48

表 2022-05-02 ResNet50 graph dlperf 运行结果

bc3af66a
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22906 MiB / 1518.46
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23036 MiB / 383.73
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23660 MiB / 1472.46
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23778 MiB / 383.53

表 2022-05-02 ResNet50 graph dlperf 运行结果

bc3af66a
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24048 MiB / 10896.8
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24128 MiB / 3046.72
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24842 MiB / 10698.48
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24936 MiB / 2946.08

表 2022-05-02 WDL ddp dlperf 运行结果

bc3af66a
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5424 MiB / 300.8107315078378
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1834 MiB / 4.106643706560135
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3550 MiB / 121.47849041596055

表 2022-05-02 WDL graph dlperf 运行结果

bc3af66a
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5130 MiB / 172.33978133276105
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1476 MiB / 1.4866517186164856
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3106 MiB / 97.79428383708

@github-actions
Copy link
Author

github-actions bot commented May 3, 2022

表 2022-05-03 ResNet50 ddp dlperf 运行结果

eb628c7
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27060 MiB / 319.35

表 2022-05-03 ResNet50 ddp dlperf 运行结果

eb628c7
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27460 MiB / 2347.76

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

eb628c7
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22906 MiB / 1510.3
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 22982 MiB / 386.16
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23674 MiB / 1505.41
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23748 MiB / 386.29

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

eb628c7
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24048 MiB / 10792.88
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24158 MiB / 3050.32
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24844 MiB / 10382.8
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24878 MiB / 3004.24

表 2022-05-03 WDL ddp dlperf 运行结果

eb628c7
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5424 MiB / 302.03404488787055
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1834 MiB / 4.141186200082302
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3550 MiB / 118.90127864480019

表 2022-05-03 WDL graph dlperf 运行结果

eb628c7
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4746 MiB / 165.72099097445607
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1476 MiB / 1.500647697597742
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 2850 MiB / 93.19385307654738

@github-actions
Copy link
Author

github-actions bot commented May 6, 2022

表 2022-05-06 ResNet50 ddp dlperf 运行结果

195162b
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27060 MiB / 317.13

表 2022-05-06 ResNet50 ddp dlperf 运行结果

195162b
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27460 MiB / 2363.28

表 2022-05-06 ResNet50 graph dlperf 运行结果

195162b
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22906 MiB / 1517.84
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23036 MiB / 389.34
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23682 MiB / 1490.49
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23718 MiB / 388.64

表 2022-05-06 ResNet50 graph dlperf 运行结果

195162b
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24048 MiB / 10727.52
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24190 MiB / 3028.24
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24810 MiB / 10649.36
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24882 MiB / 2977.76

@github-actions
Copy link
Author

github-actions bot commented May 7, 2022

表 2022-05-07 ResNet50 ddp dlperf 运行结果

0d33e09
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27060 MiB / 318.88

表 2022-05-07 ResNet50 graph dlperf 运行结果

0d33e09
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22906 MiB / 1519.96
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 22986 MiB / 388.3
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23692 MiB / 1484.33
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23694 MiB / 385.43

表 2022-05-07 WDL ddp dlperf 运行结果

0d33e09
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5424 MiB / 301.53159238398075
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1834 MiB / 4.1377574391663074
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3550 MiB / 118.44542109966278

表 2022-05-07 WDL graph dlperf 运行结果

0d33e09
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4746 MiB / 166.57868180796504
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1476 MiB / 1.4384676925837994
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3106 MiB / 96.59419701248407

@github-actions
Copy link
Author

github-actions bot commented May 8, 2022

表 2022-05-08 ResNet50 ddp dlperf 运行结果

33e604b9
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27060 MiB / 319.17

表 2022-05-08 ResNet50 ddp dlperf 运行结果

33e604b9
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27460 MiB / 2354.48

表 2022-05-08 ResNet50 graph dlperf 运行结果

33e604b9
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22906 MiB / 1520.12
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 22988 MiB / 390.22
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23670 MiB / 1484.46
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23756 MiB / 388.52

表 2022-05-08 ResNet50 graph dlperf 运行结果

33e604b9
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24050 MiB / 10861.92
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24112 MiB / 3049.44
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24814 MiB / 10251.52
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24898 MiB / 2992.16

表 2022-05-08 WDL ddp dlperf 运行结果

33e604b9
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5424 MiB / 301.1242728717625
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1834 MiB / 4.096131484955549
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3550 MiB / 116.19031031802297

表 2022-05-08 WDL graph dlperf 运行结果

33e604b9
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4746 MiB / 167.18074962496758
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1476 MiB / 1.5402142480015755
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3106 MiB / 97.48506546020508

@github-actions
Copy link
Author

github-actions bot commented May 9, 2022

表 2022-05-09 ResNet50 ddp dlperf 运行结果

4abe359b
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27460 MiB / 2358.96

表 2022-05-09 ResNet50 graph dlperf 运行结果

4abe359b
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24088 MiB / 10749.76
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24126 MiB / 3040.16
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24832 MiB / 10560.0
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24830 MiB / 3002.32

表 2022-05-09 WDL ddp dlperf 运行结果

4abe359b
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2666 MiB / 41.90006493777037
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2216 MiB / 6.55592992529273
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2436 MiB / 21.56480659171939

表 2022-05-09 WDL graph dlperf 运行结果

4abe359b
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2957 MiB / 21.70990788191557
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2191 MiB / 3.2527162805199623
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2579 MiB / 12.14893040060997

@github-actions
Copy link
Author

表 2022-05-10 ResNet50 ddp dlperf 运行结果

4abe359b
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27060 MiB / 318.63

表 2022-05-10 ResNet50 ddp dlperf 运行结果

4abe359b
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27460 MiB / 2350.16

表 2022-05-10 ResNet50 graph dlperf 运行结果

4abe359b
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22906 MiB / 1524.89
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 22986 MiB / 388.8
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23714 MiB / 1510.71
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23760 MiB / 389.59

表 2022-05-10 ResNet50 graph dlperf 运行结果

4abe359b
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24048 MiB / 10839.28
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24158 MiB / 3045.68
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24826 MiB / 10650.72
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24880 MiB / 2979.36

表 2022-05-10 WDL ddp dlperf 运行结果

4abe359b
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5424 MiB / 300.9260336570442
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1834 MiB / 4.063524655997753
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3550 MiB / 117.9467263519764

表 2022-05-10 WDL ddp dlperf 运行结果

4abe359b
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2666 MiB / 41.756586227566004
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2216 MiB / 6.7598519921302795
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2436 MiB / 21.516642469912767

表 2022-05-10 WDL graph dlperf 运行结果

4abe359b
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4236 MiB / 167.70999697223306
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1478 MiB / 1.517060823738575
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3106 MiB / 94.69545194134116

表 2022-05-10 WDL graph dlperf 运行结果

4abe359b
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2909 MiB / 21.304279651492834
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2191 MiB / 3.3603827469050884
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2579 MiB / 12.510394304990768

@github-actions
Copy link
Author

表 2022-05-11 ResNet50 ddp dlperf 运行结果

6421a51
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27060 MiB / 318.74

表 2022-05-11 ResNet50 graph dlperf 运行结果

6421a51
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22906 MiB / 1508.99
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 22984 MiB / 386.33
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23684 MiB / 1482.3
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23750 MiB / 385.08

表 2022-05-11 WDL ddp dlperf 运行结果

6421a51
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5424 MiB / 304.40418727323413
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1834 MiB / 4.070364858955145
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3550 MiB / 119.61738810315728

表 2022-05-11 WDL graph dlperf 运行结果

6421a51
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4746 MiB / 170.90625868365169
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1478 MiB / 1.4782837256789207
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3106 MiB / 93.98922028020024

@github-actions
Copy link
Author

表 2022-05-12 ResNet50 ddp dlperf 运行结果

2925f00f
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27060 MiB / 319.27

表 2022-05-12 ResNet50 ddp dlperf 运行结果

2925f00f
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27460 MiB / 2355.92

表 2022-05-12 ResNet50 graph dlperf 运行结果

2925f00f
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22906 MiB / 1515.18
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23036 MiB / 388.76
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23682 MiB / 1485.54
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23708 MiB / 386.08

表 2022-05-12 ResNet50 graph dlperf 运行结果

2925f00f
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24088 MiB / 10697.84
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24148 MiB / 3037.36
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24818 MiB / 10743.44
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24904 MiB / 2987.52

表 2022-05-12 WDL ddp dlperf 运行结果

2925f00f
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5424 MiB / 301.6199085377157
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1834 MiB / 4.097774285823107
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3550 MiB / 119.3243205472827

表 2022-05-12 WDL ddp dlperf 运行结果

2925f00f
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2666 MiB / 41.96615158393979
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2216 MiB / 6.595474328845739
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2436 MiB / 21.248769514262676

表 2022-05-12 WDL graph dlperf 运行结果

2925f00f
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4746 MiB / 171.79899736493826
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1476 MiB / 1.4992695860564709
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3106 MiB / 96.29827499762177

表 2022-05-12 WDL graph dlperf 运行结果

2925f00f
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2973 MiB / 21.85895549878478
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2191 MiB / 3.3689139559865
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2571 MiB / 12.45945905894041

@github-actions
Copy link
Author

表 2022-05-13 ResNet50 ddp dlperf 运行结果

0ad6733
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27460 MiB / 2363.6

表 2022-05-13 ResNet50 graph dlperf 运行结果

0ad6733
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24090 MiB / 10815.2
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24148 MiB / 3036.96
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24858 MiB / 10771.04
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24874 MiB / 2966.72

表 2022-05-13 WDL ddp dlperf 运行结果

0ad6733
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2666 MiB / 40.831202775239944
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2216 MiB / 6.6353478245437145
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2436 MiB / 21.402670316398144

表 2022-05-13 WDL graph dlperf 运行结果

0ad6733
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2973 MiB / 21.326240353286266
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2191 MiB / 3.2991334311664104
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2547 MiB / 12.27147825807333

表 2022-05-13 ResNet50 ddp dlperf 运行结果

8ae6c41
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27060 MiB / 319.34

表 2022-05-13 ResNet50 graph dlperf 运行结果

8ae6c41
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22906 MiB / 1512.34
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 22986 MiB / 384.34
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23678 MiB / 1484.41
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23698 MiB / 383.69

表 2022-05-13 WDL ddp dlperf 运行结果

8ae6c41
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5424 MiB / 318.83345828577876
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1834 MiB / 4.313508257269859
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3550 MiB / 137.03867534920573

表 2022-05-13 WDL graph dlperf 运行结果

8ae6c41
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4746 MiB / 158.39239247515798
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1476 MiB / 1.4626655206084251
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3106 MiB / 84.34315771237016

@github-actions
Copy link
Author

表 2022-05-14 ResNet50 ddp dlperf 运行结果

944ad628
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27060 MiB / 319.29

表 2022-05-14 ResNet50 ddp dlperf 运行结果

944ad628
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27460 MiB / 2359.52

表 2022-05-14 ResNet50 graph dlperf 运行结果

944ad628
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22906 MiB / 1512.03
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 22982 MiB / 392.03
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23692 MiB / 1503.81
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23764 MiB / 385.77

表 2022-05-14 ResNet50 graph dlperf 运行结果

944ad628
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24088 MiB / 10720.48
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24144 MiB / 3029.36
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24846 MiB / 10729.28
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24852 MiB / 2971.04

表 2022-05-14 WDL ddp dlperf 运行结果

944ad628
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5424 MiB / 301.72775399684906
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1834 MiB / 4.147641833871603
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3550 MiB / 120.90900176018476

表 2022-05-14 WDL graph dlperf 运行结果

944ad628
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4746 MiB / 170.40045065060258
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1476 MiB / 1.5055376626551151
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3042 MiB / 97.98973647132516

@github-actions
Copy link
Author

表 2022-05-15 ResNet50 ddp dlperf 运行结果

3042749
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27060 MiB / 319.08

表 2022-05-15 ResNet50 graph dlperf 运行结果

3042749
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22906 MiB / 1509.87
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 22986 MiB / 389.5
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23688 MiB / 1510.65
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23708 MiB / 389.6

表 2022-05-15 WDL ddp dlperf 运行结果

3042749
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5424 MiB / 300.9220148399472
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1834 MiB / 4.082046613097191
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3550 MiB / 118.52849021926522

表 2022-05-15 WDL ddp dlperf 运行结果

3042749
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2666 MiB / 40.78918404132128
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n8g 2216 MiB / 6.6182686649262905
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2436 MiB / 21.24025771766901

表 2022-05-15 WDL graph dlperf 运行结果

3042749
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4746 MiB / 166.11498599499464
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1476 MiB / 1.4970311857759953
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3106 MiB / 96.58953167870641

表 2022-05-15 WDL graph dlperf 运行结果

3042749
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n8g 2973 MiB / 21.392423320561647
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n8g 2191 MiB / 3.225696075707674
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n8g 2579 MiB / 12.137069839984179

@github-actions
Copy link
Author

表 2022-05-16 ResNet50 ddp dlperf 运行结果

4787170d0
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27060 MiB / 318.83

表 2022-05-16 ResNet50 graph dlperf 运行结果

4787170d0
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22906 MiB / 1519.44
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 22984 MiB / 387.26
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23704 MiB / 1493.92
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23788 MiB / 383.87

表 2022-05-16 WDL ddp dlperf 运行结果

4787170d0
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5424 MiB / 301.2004911378026
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1834 MiB / 4.154437091201544
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3550 MiB / 117.7886158414185

表 2022-05-16 WDL graph dlperf 运行结果

4787170d0
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4236 MiB / 166.5742148309946
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1476 MiB / 1.518830120563507
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3106 MiB / 94.697659868747

表 2022-05-16 ResNet50 ddp dlperf 运行结果

5e6451ca7
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27460 MiB / 2351.6

表 2022-05-16 ResNet50 graph dlperf 运行结果

5e6451ca7
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24068 MiB / 10794.08
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24160 MiB / 3027.04
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24890 MiB / 10768.32
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24920 MiB / 2988.32

@github-actions
Copy link
Author

表 2022-05-17 WDL graph dlperf 运行结果

a2145c727
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 2022-05-17 ResNet50 ddp dlperf 运行结果

cdf34dbc2
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g None MiB / 0

表 2022-05-17 ResNet50 graph dlperf 运行结果

cdf34dbc2
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g None MiB / 0
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g None MiB / 0
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g None MiB / 0
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g None MiB / 0

@github-actions
Copy link
Author

表 2022-05-20 ResNet50 ddp dlperf 运行结果

ad9aa8c
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27066 MiB / 321.87

表 2022-05-20 ResNet50 graph dlperf 运行结果

ad9aa8c
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22912 MiB / 1540.91
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23038 MiB / 394.09
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23694 MiB / 1520.75
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23724 MiB / 395.84

表 2022-05-20 WDL ddp dlperf 运行结果

ad9aa8c
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5430 MiB / 312.14761059358716
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1840 MiB / 4.187230002135038
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3556 MiB / 130.22070188820362

表 2022-05-20 WDL graph dlperf 运行结果

ad9aa8c
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4752 MiB / 153.16442050412297
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1482 MiB / 1.455139085650444
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3048 MiB / 80.48255839571357

@github-actions
Copy link
Author

表 2022-05-21 ResNet50 graph train 50E运行结果

0ed4f466
Memory/Throughput/TOP1
ResNet50_graph_train_gpudecode_FP16_b192_1n8g 11410 MiB / 10630.08 / 76.0

表 2022-05-21 ResNet50 ddp dlperf 运行结果

38b6495
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27066 MiB / 318.33

表 2022-05-21 ResNet50 ddp dlperf 运行结果

38b6495
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27466 MiB / 2356.08

表 2022-05-21 ResNet50 graph dlperf 运行结果

38b6495
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22912 MiB / 1509.46
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23014 MiB / 387.77
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23700 MiB / 1496.67
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23746 MiB / 385.71

表 2022-05-21 ResNet50 graph dlperf 运行结果

38b6495
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24074 MiB / 10746.0
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24190 MiB / 3041.92
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24846 MiB / 10832.24
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24890 MiB / 2970.0

表 2022-05-21 WDL ddp dlperf 运行结果

38b6495
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5430 MiB / 309.59268153458834
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1840 MiB / 4.2023018412292
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3556 MiB / 129.66127562895417

表 2022-05-21 WDL graph dlperf 运行结果

38b6495
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4752 MiB / 165.80314771458507
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1482 MiB / 1.457510706037283
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3112 MiB / 85.30815675109625

@github-actions
Copy link
Author

表 2022-05-22 ResNet50 ddp dlperf 运行结果

0afc09f
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27066 MiB / 318.75

表 2022-05-22 ResNet50 graph dlperf 运行结果

0afc09f
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22912 MiB / 1517.78
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 22990 MiB / 387.46
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23688 MiB / 1490.34
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23726 MiB / 387.15

表 2022-05-22 WDL ddp dlperf 运行结果

0afc09f
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5430 MiB / 302.3267007730901
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1840 MiB / 4.155577626079321
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3556 MiB / 119.75239384546876

表 2022-05-22 WDL graph dlperf 运行结果

0afc09f
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4752 MiB / 167.89393467083573
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1484 MiB / 1.4240652769804
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3112 MiB / 89.4975531771779

表 2022-05-22 ResNet50 ddp dlperf 运行结果

7476de9
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27466 MiB / 2350.4

表 2022-05-22 ResNet50 graph dlperf 运行结果

7476de9
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24074 MiB / 10817.04
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24148 MiB / 3041.44
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24878 MiB / 10727.84
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24912 MiB / 3001.92

@github-actions
Copy link
Author

表 2022-05-23 ResNet50 ddp dlperf 运行结果

7476de9
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27066 MiB / 318.9

表 2022-05-23 ResNet50 ddp dlperf 运行结果

7476de9
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27466 MiB / 2346.88

表 2022-05-23 ResNet50 graph dlperf 运行结果

7476de9
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22912 MiB / 1510.76
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23042 MiB / 387.92
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23702 MiB / 1496.37
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23716 MiB / 387.02

表 2022-05-23 ResNet50 graph dlperf 运行结果

7476de9
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24074 MiB / 10702.24
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24154 MiB / 3029.44
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24842 MiB / 10823.36
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24900 MiB / 2942.4

表 2022-05-23 WDL ddp dlperf 运行结果

7476de9
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5430 MiB / 304.2198572233319
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1840 MiB / 4.146246988326311
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3556 MiB / 120.27159281075001

表 2022-05-23 WDL graph dlperf 运行结果

7476de9
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4752 MiB / 168.9178863465786
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1482 MiB / 1.5098589770495892
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 2856 MiB / 94.29860880225897

@github-actions
Copy link
Author

表 2022-05-26 ResNet50 ddp dlperf 运行结果

3bf33ad
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27066 MiB / 318.01

表 2022-05-26 ResNet50 graph dlperf 运行结果

3bf33ad
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22912 MiB / 1520.92
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23066 MiB / 387.07
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23690 MiB / 1504.9
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23712 MiB / 387.3

表 2022-05-26 WDL ddp dlperf 运行结果

3bf33ad
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5430 MiB / 304.9048557430506
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1840 MiB / 4.050893165171146
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3556 MiB / 127.47526326775551

表 2022-05-26 WDL graph dlperf 运行结果

3bf33ad
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4752 MiB / 168.9307601414621
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1482 MiB / 1.4917825795710087
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 2856 MiB / 90.50108759850264

表 2022-05-26 ResNet50 ddp dlperf 运行结果

dd83698
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n8g 27466 MiB / 2341.44

表 2022-05-26 ResNet50 graph dlperf 运行结果

dd83698
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n8g 24054 MiB / 10763.36
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n8g 24156 MiB / 3041.68
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n8g 24850 MiB / 11202.72
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n8g 24910 MiB / 3016.96

@github-actions
Copy link
Author

表 2022-05-27 ResNet50 ddp dlperf 运行结果

a447420
Memory/Throughput
ResNet50_ddp_dlperf_cpudecode_FP32_b256_1n1g 27066 MiB / 318.91

表 2022-05-27 ResNet50 graph dlperf 运行结果

a447420
Memory/Throughput
ResNet50_graph_dlperf_cpudecode_FP16_b512_1n1g 22912 MiB / 1508.85
ResNet50_graph_dlperf_cpudecode_FP32_b256_1n1g 23018 MiB / 390.03
ResNet50_graph_dlperf_gpudecode_FP16_b512_1n1g 23690 MiB / 1498.06
ResNet50_graph_dlperf_gpudecode_FP32_b256_1n1g 23740 MiB / 391.05

表 2022-05-27 WDL ddp dlperf 运行结果

a447420
Memory/latency(ms)
WDL_ddp_dlperf_vs2322444_devs16_hun2_b131072_1n1g 5430 MiB / 300.89791110903025
WDL_ddp_dlperf_vs2322444_devs16_hun2_b512_1n1g 1840 MiB / 4.119891092181206
WDL_ddp_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3556 MiB / 120.54904874786735

表 2022-05-27 WDL graph dlperf 运行结果

a447420
Memory/latency(ms)
WDL_graph_dlperf_vs2322444_devs16_hun2_b131072_1n1g 4752 MiB / 167.497763376683
WDL_graph_dlperf_vs2322444_devs16_hun2_b512_1n1g 1482 MiB / 1.4924899265170097
WDL_graph_dlperf_vs2322444_devs16_hun2_b65536_1n1g 3112 MiB / 92.99978451058269

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