Skip to content

Commit 99150c4

Browse files
lxwlaqlxwlaq
authored
update performance data, test=document_fix (#8928)
1 parent 2a6b0b9 commit 99150c4

File tree

1 file changed

+65
-30
lines changed

1 file changed

+65
-30
lines changed

‎docs/performance/benchmark.md

Lines changed: 65 additions & 30 deletions
Original file line numberDiff line numberDiff line change
@@ -64,8 +64,8 @@
6464
* GPU: 4 core Mali-T860
6565

6666
* 测试说明
67-
* Branch: release/v2.10, commit id: b2e9776
68-
* 使用 Android ndk-r20b，armv8 编译
67+
* Branch: release/v2.11, commit id: 4a3bdbe
68+
* 使用 Android ndk-r22b，armv7 armv8 编译
6969
* CPU 线程数设为 1，绑定大核
7070
* 在 GPU 上运行时，开启了 Auto Tune
7171
* warmup=20, repeats=600，统计平均时间，单位 ms
@@ -74,58 +74,93 @@
7474

7575
### fp32 浮点模型测试数据
7676

77-
#### CPU 数据
77+
#### ARMV8 CPU 数据
7878
运行时精度为 fp32 的性能数据如下：
7979

8080
|模型|骁龙 865|骁龙 855|骁龙 845|骁龙 835|骁龙 625|麒麟 990|麒麟 980|RK3399|
8181
|:----|----:|----:|----:|----:|----:|----:|----:|----:|
82-
|MobileNetV1|28.52 |29.22 |60.78 |82.54 |144.20 |38.16 |32.86 |111.76 |
83-
|MobileNetV2|18.84 |23.17 |42.74 |56.23 |107.66 |24.91 |22.51 |79.95 |
84-
|MobileNetV3_large_x1_0|14.55 |18.39 |32.49 |41.95 |96.30 |19.46 |17.78 |71.17 |
85-
|MobileNetV3_small_x1_0|4.75 |6.41 |9.98 |13.98 |37.99 |6.50 |6.00 |23.34 |
86-
|ResNet50|162.40 |192.88 |430.72 |490.54 |842.96 |221.81 |191.14 |638.29 |
87-
|SSD_MobileNetV3_large|33.87 |42.84 |84.70 |103.32 |199.60 |46.02 |40.95 |157.08 |
88-
|HRNet_w18|640.62 |835.62 |1687.78 |2048.81 |4724.20 |910.09 |820.42 |3380.08 |
82+
|MobileNetV1|28.38 |29.04 |58.70 |82.48 |143.64 |38.22 |32.50 |103.39 |
83+
|MobileNetV2|18.52 |19.37 |37.61 |53.47 |106.17 |24.93 |22.07 |73.37 |
84+
|MobileNetV3_large_x1_0|14.46 |15.59 |29.55 |41.74 |96.11 |19.38 |17.46 |63.34 |
85+
|MobileNetV3_small_x1_0|4.73 |5.39 |9.76 |13.45 |39.08 |6.41 |5.85 |21.67 |
86+
|ResNet50|160.97 |161.13 |339.59 |484.79 |831.62 |222.21 |190.38 |616.68 |
87+
|SSD_MobileNetV3_large|33.62 |36.07 |69.76 |99.88 |193.79 |46.95 |40.87 |153.94 |
88+
|HRNet_w18|645.30 |694.41 |1395.66 |2063.99 |4717.07 |905.07 |792.17 |3491.22 |
89+
90+
运行时精度为 fp16 的性能数据如下：
91+
92+
|模型|骁龙 865|骁龙 855|骁龙 845|麒麟 990|
93+
|:----|----:|----:|----:|----:|
94+
|MobileNetV1|14.83 |15.79 |29.62 |20.69 |
95+
|MobileNetV2|9.31 |10.29 |19.10 |12.34 |
96+
|MobileNetV3_large_x1_0|7.63 |8.15 |15.77 |9.63 |
97+
|MobileNetV3_small_x1_0|2.48 |2.90 |5.64 |3.36 |
98+
|ResNet50|82.00 |83.87 |166.00 |106.65 |
99+
|SSD_MobileNetV3_large|17.78 |19.40 |36.29 |23.62 |
100+
|HRNet_w18|377.88 |418.99 |812.44 |529.34 |
101+
102+
#### ARMV7 CPU 数据
103+
运行时精度为 fp32 的性能数据如下：
104+
105+
|模型|骁龙 865|骁龙 855|骁龙 845|骁龙 835|骁龙 625|麒麟 990|麒麟 980|RK3399|
106+
|:----|----:|----:|----:|----:|----:|----:|----:|----:|
107+
|MobileNetV1|31.21 |33.44 |68.04 |93.72 |147.85 |41.18 |35.49 |116.44 |
108+
|MobileNetV2|21.47 |25.79 |46.19 |64.54 |131.05 |28.18 |25.08 |84.09 |
109+
|MobileNetV3_large_x1_0|16.81 |19.47 |34.80 |48.61 |102.01 |21.96 |19.75 |66.28 |
110+
|MobileNetV3_small_x1_0|5.44 |6.48 |11.25 |15.44 |37.74 |7.28 |6.73 |21.01 |
111+
|ResNet50|177.80 |188.07 |377.67 |519.61 |886.98 |237.47 |203.60 |657.27 |
112+
|SSD_MobileNetV3_large|38.02 |45.21 |82.34 |115.40 |210.73 |52.26 |46.02 |156.85 |
113+
|HRNet_w18|733.96 |839.54 |1532.35 |2200.64 |5029.30 |989.71 |869.29 |3256.20 |
89114

90115

91116
运行时精度为 fp16 的性能数据如下：
92117

93118
|模型|骁龙 865|骁龙 855|骁龙 845|麒麟 990|
94119
|:----|----:|----:|----:|----:|
95-
|MobileNetV1|14.83 |15.79 |29.62 |20.64 |
96-
|MobileNetV2|9.49 |10.28 |18.93 |12.29 |
97-
|MobileNetV3_large_x1_0|7.84 |8.29 |16.00 |9.75 |
98-
|MobileNetV3_small_x1_0|2.58 |3.03 |5.85 |3.47 |
99-
|ResNet50|84.06 |87.10 |179.46 |109.38 |
100-
|SSD_MobileNetV3_large|18.32 |19.99 |40.13 |24.37 |
101-
|HRNet_w18|388.43 |430.27 |954.59 |544.75 |
120+
|MobileNetV1|15.81 |16.39 |58.72 |21.04 |
121+
|MobileNetV2|10.59 |11.55 |25.28 |13.96 |
122+
|MobileNetV3_large_x1_0|8.69 |9.50 |20.01 |11.00 |
123+
|MobileNetV3_small_x1_0|3.08 |3.55 |6.77 |4.00 |
124+
|ResNet50|86.62 |89.15 |289.57 |110.00 |
125+
|SSD_MobileNetV3_large|20.03 |22.12 |48.93 |27.03 |
126+
|HRNet_w18|481.79 |520.02 |1032.48 |650.79 |
102127

103128

104129
#### GPU 数据
105130

106131
|模型|骁龙 865|骁龙 855|骁龙 845|骁龙 835|骁龙 625|麒麟 990|麒麟 980|RK3399|
107132
|:----|----:|----:|----:|----:|----:|----:|----:|----:|
108-
|MobileNetV1|7.05 |8.85 |10.46 |10.87 |71.42 |8.15 |13.74 |45.91 |
109-
|MobileNetV2|9.48 |9.70 |8.58 |14.14 |52.09 |9.32 |13.08 |37.27 |
110-
|MobileNetV3_large_x1_0|8.90 |9.11 |10.20 |12.04 |46.48 |9.81 |15.19 |32.92 |
111-
|MobileNetV3_small_x1_0|5.79 |5.54 |8.52 |11.43 |20.00 |6.45 |8.71 |18.42 |
112-
|ResNet50|29.70 |35.46 |45.23 |53.66 |392.62 |36.15 |54.23 |238.12 |
113-
|SSD_MobileNetV3_large|27.69 |35.21 |33.37 |42.31 |152.37 |27.25 |35.79 |90.37 |
133+
|MobileNetV1|6.39 |6.41 |8.26 |10.13 |56.04 |8.18 |15.28 |45.94 |
134+
|MobileNetV2|8.96 |8.07 |7.96 |11.49 |43.65 |9.07 |15.83 |37.67 |
135+
|MobileNetV3_large_x1_0|9.89 |9.30 |11.29 |16.64 |40.58 |9.43 |14.89 |32.57 |
136+
|MobileNetV3_small_x1_0|8.10 |6.79 |10.13 |10.37 |19.12 |6.59 |7.69 |18.63 |
137+
|ResNet50|25.22 |29.70 |39.11 |47.65 |337.10 |36.90 |55.54 |237.41 |
138+
|SSD_MobileNetV3_large|22.68 |30.21 |26.58 |26.53 |129.34 |25.14 |32.64 |102.15 |
114139

115140

116141
### int8 量化模型测试数据
117142

118-
#### CPU 数据
143+
#### ARMV8 CPU 数据
119144

120145
|模型|骁龙 865|骁龙 855|骁龙 845|骁龙 835|骁龙 625|麒麟 990|麒麟 980|RK3399|
121146
|:----|----:|----:|----:|----:|----:|----:|----:|----:|
122-
|MobileNetV1_quant|11.34 |14.81 |52.34 |55.69 |118.76 |14.80 |13.83 |78.30 |
123-
|MobileNetV2_quant|10.55 |14.06 |33.99 |40.87 |85.81 |14.22 |13.06 |57.94 |
124-
|MobileNetV3_large_x1_0_quant|8.11 |10.76 |24.63 |31.30 |70.86 |10.52 |9.73 |48.36 |
125-
|MobileNetV3_small_x1_0_quant|3.04 |4.20 |8.93 |11.27 |25.13 |4.10 |3.75 |17.87 |
126-
|ResNet50_quant|64.60 |80.46 |313.63 |331.30 |691.06 |81.65 |74.68 |489.30 |
127-
|SSD_MobileNetV3_large_quant|20.84 |22.82 |64.16 |74.12 |165.91 |27.11 |25.29 |119.92 |
147+
|MobileNetV1_quant|11.18 |11.86 |44.12 |55.57 |117.41 |14.75 |13.58 |77.60 |
148+
|MobileNetV2_quant|10.37 |11.38 |33.45 |40.78 |83.47 |14.14 |12.94 |56.66 |
149+
|MobileNetV3_large_x1_0_quant|8.03 |8.88 |24.32 |30.57 |66.76 |10.41 |9.57 |46.66 |
150+
|MobileNetV3_small_x1_0_quant|2.89 |3.29 |8.41 |10.86 |22.83 |4.23 |3.90 |15.83 |
151+
|ResNet50_quant|64.50 |66.81 |263.17 |331.76 |690.01 |82.07 |74.52 |473.00 |
152+
|SSD_MobileNetV3_large_quant|20.43 |22.71 |57.45 |73.97 |164.65 |27.26 |25.08 |114.71 |
128153

154+
#### ARMV7 CPU 数据
155+
156+
|模型|骁龙 865|骁龙 855|骁龙 845|骁龙 835|骁龙 625|麒麟 990|麒麟 980|RK3399|
157+
|:----|----:|----:|----:|----:|----:|----:|----:|----:|
158+
|MobileNetV1_quant|15.26 |16.11 |53.71 |63.43 |129.72 |19.86 |17.97 |85.89 |
159+
|MobileNetV2_quant|16.44 |17.96 |43.67 |53.32 |105.39 |22.08 |19.90 |72.77 |
160+
|MobileNetV3_large_x1_0_quant|11.20 |12.50 |30.72 |37.52 |76.98 |15.19 |13.99 |51.93 |
161+
|MobileNetV3_small_x1_0_quant|4.27 |4.82 |10.58 |13.23 |26.86 |5.98 |5.51 |18.55 |
162+
|ResNet50_quant|75.82 |83.01 |301.21 |360.67 |763.98 |96.03 |85.56 |504.01 |
163+
|SSD_MobileNetV3_large_quant|27.64 |31.40 |73.19 |92.01 |194.78 |36.87 |33.25 |131.47 |
129164

130165
## 华为昇腾 NPU 的性能数据
131166
请参考 [Paddle Lite 使用华为昇腾 NPU 预测部署](../demo_guides/huawei_ascend_npu)的最新性能数据

0 commit comments

Comments
 (0)