search

BUG1989 / ncnn-benchmark

The benchmark of ncnn that is a high-performance neural network inference framework optimized for the mobile platform
72 stars 19 forks source link

confused about performance on 8xA53 #1

Closed wl3b10s closed 6 years ago

wl3b10s commented 6 years ago

i made ncnn with default option and use your benchmark test of squeezenet on 8*A53 CPU in ubuntu 16.04.

i got about 960ms with about 700% CPU usage which means nearly full use of 8 core.

this result is much far behind your performance on A17*4 platform.

do you have any suggestion or advice on what's wrong for my test?

BUG1989 commented 6 years ago
  1. This repo lib does not support arm linux system platform,which ncnn-lib do you used?
  2. If you use the arm platform,please make sure that using the ncnn-android lib work on android platform;
  3. Please try to use single model and report it's timeconsuming result;
  4. Cortex-A53 is slower than Cortex-A17 in the same freq;
  5. Good luck!