Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Deprost__Emiel
Beginner
54 Views

Bandwidth information on Myriad devices

Hello,

With the NCSDK  there is the mvNCProfile tool that allows to the execution time of every layer as well as the bandwidth and MFLOPS. 

The closest equivalent that I  have found in the OpenVINO toolkit is the get_perf_counts() method. But this method only shows the time for each layer.

Is there any way to see the bandwidth information with the OpenVINO toolkit?

 

Thank you in advance

0 Kudos
2 Replies
Shubha_R_Intel
Employee
54 Views

Dear Emiel, you should transition away from NCSDK and toward OpenVino. There is myriad_perfcheck.exe found here within OpenVIno R5:

deployment_tools\inference_engine\bin\intel64\Release\

There is a recent forum post on this very topic here:

https://software.intel.com/en-us/forums/computer-vision/topic/806035

Please peruse the generic perfcheck documentation here:

https://docs.openvinotoolkit.org/R5/_samples_perfcheck_README.html

perfcheck gives you Min, Max and Avg FPS (Frames Per Second) and execution time which is not like the  mvNCProfile tool that allows to the execution time of every layer as well as the bandwidth and MFLOPS. I will check on an mvNCProfile equivalent within OpenVino and report my findings here.

Thank you !

Shubha

 

Shubha_R_Intel
Employee
54 Views

Dear Emiel:

Please  wait a bit for the 2019-R1 release coming soon. In this release we have improved the benchmark_app tool considerably. It can report per-layer performance counters collected including execution time. The possibility also exists to collect performance statistics using a variety of interesting and convenient report types.

Also, you can check some existing samples/demos which can accomodate your particular model. Virtually all of the samples support PerformanceCounters (mostly with “–pc” option passed to the sample). 

But regarding MFLOPS – this performance metric is not available today within OpenVino.

Hope it helps, and thank you for using OpenVino !

Shubha