Hi,
Greetings.
I am using snpe-1.8.0 SDK. I am trying to run inception_v3 model shifted with sdk on Android(snapdragon 820 mini ADP Developement Kit)
and Linux Host(Ubuntu).
I can see huge difference in time snpe-net-run took to run the model on the two platforms.
Using GPU Runtime on Linux Host snpe-net-run executed the model (inception_v3.dlc) in 0.428seconds, but on Android Device same model is taking 2.65seconds.
*/ I am wondering what might be the reason for this difference .
please help.
Thanks,
Kshama Vete
*/
Hi,
You stated in your query that execution of "snpe-net-run" is faster in Linux machine compared to Android device.
To agree to this statement, we have to confirm if both GPU's are of the same configuration. If it not so, the execution speed of configuration device is faster.
Assuming both the GPU's are of the same configurations, the next thing we have to confirm is, whether processing time is shared with another process along with snpe-net-run. If processing time is shared, we have to consider only the running time of the process (in our case snpe-net-run).
Kindly refer below link for Benchmarking and Accuracy,
https://developer.qualcomm.com/docs/snpe/benchmarking.html