Performance query for eiq-sample-apps on imx8m nano cpu vs gpu

[tyler_ward]

I am atempting to confirm my performance numbers when running the eiq demos on the imx8m nano.

I am running a stock image (fsl-image-validation-imx) using the 4.14.98-2.3.0 yocto sumo manifest in the imx8mnevk board. I am using the opencv obj_recognition_cpp example code modified to set the Preferable target either set to CPU or OPENCL

edit to force hardware openCL shown below

  //Net net = readNetFromTensorflow(model, config);
  net = readNetFromCaffe(config, model);
  net.setPreferableTarget(cv::dnn::DNN_TARGET_OPENCL);
}

When using cpu processing I get an Inference fps of about 3 fps

When using the gpu for open cl I get an Inference fps of arround 1/12 (0.08) fps 

Are these the expected performance levels for running open cl on the gpu or have I misconfigured something?

Thanks

Tyler

[Bio_TICFSL]

Hello Tyler,

1. you can follow chapter 15 Application Programing Recommandations in the  iMX_Graphics_User's_Guide to optimize the apps.

2. Customer can use vProfiler and vAnalyzer ( VivanteVTK) to get a more detailed GPU performance, including GPU utilization, pixel rate, GPU idle cycles, and so on. For more info, please refer to vProfiler and vAnalyzer in the Graphics User Guide documentation.

 When GPU load is very high, other GPU apps will be slower. So, users must balance GPU resource among different GPU apps.

Regards