Low Performance for AlexNet Inference on big cores

My environment is odroid-xu3 that constructed with exynos5422 Cortex-A15 2.0Ghz quad core and Cortex-A7 quad core CPUs and have Ubuntu 18.04, Linux kernel 4.14.141-69.

My issue is that I am running AlexNet example from the compute library and setting the cores to the big A15 cores using taskset and threads=4, but the execution time of the inference, which I measure using the time command is coming out to be around :

real 0m11.211s
user 0m9.086s
sys 0m2.041s

Isnt't this too high? I have read papers and they have reported inference time in the order of mili seconds. What can be the issue?

I built the library using assert=0 and debug =0 NEON =1 OPENCL=1

and the command I am using to run the example is 

taskset -c 4,5,6,7 ./build/examples/graph_alexnet --data=$PATH_ASSETS --image=$PATH_ASSETS/go_kart.ppm --labels=$PATH_ASSETS/labels.txt


Parents Reply Children
No data
More questions in this forum