Next up we have a couple of
AI-related benchmarks, AIMark and AiTuTu. Both of these benchmarks are available publicly, but are still in beta. AImark uses the Resnet34, Inception v3, MobileNet-SSD, and Deeplab v3+ models for image classification, image recognition, and image segmentation tasks and determines AI performance of a device by judging recognition efficiency, before outputting an overall score and individual scores and frame rates for a particular test routine.
 |
|
AIMark |
| AI & Machine Learning Benchmarks |
|
The Snapdragon 865 significantly outpaces every other device we tested in AIMark, across the board, regardless of the workload. Keep in mind when looking at these results, that the Snapdragon 865 has dedicated hardware for accelerating AI workloads and this benchmark has been updated to support the latest Snapdragon Neural Processing Engine (SNPE) SDK, as well as an asynchronous compute mode offered by the Snapdragon 865.
|
|
AITuTu |
|
AI & Machine Learning Benchmarks |
|
AITuTu evaluates the AI performance of a device by leveraging two mainstream neural network models, Inception v3 for Image Classification and MobileNet-SSD for Object Detection. The benchmark determines the device’s accuracy and speed when inferring data in each test.
Accuracy is comparable across all of the devices we tested in AITuTu, but the Snapdragon 865's speed and overall performance crush the previous-gen Snapdragon 855-based Samsung
Galaxy Note 10.
