Geekbench ML Score

292
TensorFlow Lite CPU Inference Score
Geekbench ML 0.6.0 for Android AArch64

Result Information

Upload Date June 11 2024 07:13 PM
Views 1

System Information

System Information
Operating System Android 14
Model samsung SM-A546E
Model IDsamsung SM-A546E
Motherboard s5e8835
Governorenergy_aware
CPU Information
Name ARM ARMv8
Topology 1 Processor, 8 Cores
Identifier ARM implementer 65 architecture 8 variant 1 part 3393 revision 1
Base Frequency 2.00 GHz
Cluster 14 Cores @ 2.00 GHz
Cluster 24 Cores @ 2.40 GHz
Memory Information
Size 7.26 GB
Inference Information
Framework TensorFlow Lite
Backend CPU
Device Default

Inference Performance

Workload Accuracy Score
Image Classification (F32)
100% 142
26.5 IPS
 
Image Classification (F16)
100% 186
34.8 IPS
 
Image Classification (I8)
97% 204
38.1 IPS
 
Image Segmentation (F32)
100% 208
3.47 IPS
 
Image Segmentation (F16)
100% 222
3.70 IPS
 
Image Segmentation (I8)
98% 296
4.95 IPS
 
Pose Estimation (F32)
100% 425
0.51 IPS
 
Pose Estimation (F16)
100% 421
0.51 IPS
 
Pose Estimation (I8)
100% 2288
2.77 IPS
 
Object Detection (F32)
100% 155
11.6 IPS
 
Object Detection (F16)
100% 86
6.44 IPS
 
Object Detection (I8)
61% 186
13.9 IPS
 
Face Detection (F32)
100% 293
3.49 IPS
 
Face Detection (F16)
100% 297
3.53 IPS
 
Face Detection (I8)
86% 440
5.24 IPS
 
Depth Estimation (F32)
100% 298
2.31 IPS
 
Depth Estimation (F16)
100% 353
2.74 IPS
 
Depth Estimation (I8)
95% 434
3.37 IPS
 
Style Transfer (F32)
100% 752
0.99 IPS
 
Style Transfer (F16)
100% 679
0.89 IPS
 
Style Transfer (I8)
98% 1597
2.10 IPS
 
Image Super-Resolution (F32)
100% 141
5.03 IPS
 
Image Super-Resolution (F16)
100% 138
4.93 IPS
 
Image Super-Resolution (I8)
98% 232
8.30 IPS
 
Text Classification (F32)
100% 272
390.6 IPS
 
Text Classification (F16)
100% 227
326.9 IPS
 
Text Classification (I8)
92% 306
439.3 IPS
 
Machine Translation (F32)
100% 317
5.83 IPS
 
Machine Translation (F16)
100% 333
6.13 IPS
 
Machine Translation (I8)
64% 101
1.86 IPS
 

Compare

Set Baseline Geekbench 6 now available