I am working on Hikey970 board with ARM Mali-G72 MP12 GPU. I am using ARM Compute Library and following repo mentioned below:
I have executed 5 networks and observed inference times on CPU bit core, CPU LITTLE core and GPU. For ResNet50, the CPU LITTLE core outperforms both big core and GPU. I tried using specific core fully for ResNet50 only, and still the inference time is more in big core. Could you guys help me understand why ResNet50 shows such behavior?
Same Question: https://community.arm.com/developer/ip-products/processors/f/cortex-a-forum/49141/why-performance-is-higher-on-little-cores ?