I start GPA tool and use system analyzer to view total GPU load and mfx usage. Then I start sample_encode provided in Media Studio Professional eval version to encode. The sample_encode.exe hangs at first frame itself at "Processing Frame: 1"
If I start GPA tool after starting sample_encode.exe, sample_encode.exe runs fine till GPA system analyzer is started, once system analyzer is started, sample_encode.exe hangs at stating for e.g. "Processing Frame: 400".
If I don't use GPA tool at-all, sample_encode.exe runs to completion.
I am running this on Win-10, Core-M5 Intel stick
Unfortunately, GPA is not supported on the Intel sticks. However, I am trying to see if I can replicate your problem on a desktop system. Are you able to use a desktop or laptop to see if the problem still occurs?
You say you are trying to use the sample_encode that comes with Media Studio Professional - do you have a link to share where you downloaded this from? I want to be sure I am using exatly what you are using.
Also, for my understandling, what are you trying to accomplish by running System Analyzer while running sample_encode? Perhaps there is another way to acomplish what you are aiming to do.
Intel Media Server can be downloaded from https://software.intel.com/en-us/intel-media-server-studio/try-buy
Sample code for this can be downloaded from https://software.intel.com/en-us/intel-media-server-studio-support/code-samples
I need to see GPU load and what fraction of GPU is used for mfx (Media SDK related calls). Let me know if there is an alternate way to find this.
I installed and the sample_encoder you linked to and ran it with and without System Analyzer and the encoder finishes each time. The sample that comes with the pack is small, so it may be finishing too quickly since the sample is so short which leaves me a very small amount of time to start SA after I start encoding the sample.
I am running on a Win10 desktop machine with an i7 and 16GB of ram.
Are you encoding the sample YUV file that comes with the install or a different file?
Are you able to use SA on the Intel Stick with other applications? If you are running SA on a machine with Intel gfx you will be able to get the GPU mfx metrics for an application.
I am encoding a different file which has approx 800-900 frames captured from camera in YUV420 format. BTW, I am able to run the SA with sample_encode on my laptop which is Core i5 with same YUV file and same version of Media Server and sample_encode with SA on.
That's great news that you can encode the video with SA running on your laptop!
Are you able to see/get the metrics you are looking for from the laptop?
If I am understanding you correctly, the Intel Stick looks like it could be the issue we have been discussing. Correct?
Yes, it seems to be problem with stick. However, our target platform is Intel Stick, hence we need to find the GPU usage on Intel stick. If you could suggest some other tool or method to find GPU usage, we can use that. Or any API that we can call from our application to check GPU usage.
I see. Are you able to put the Intel Stick and another computer, like your laptop, on to the same network? If so, you can use the laptop to remotely monitor the Intel Stick with 'Network Usage Mode'. More info can be found here: https://software.intel.com/en-us/node/597178.
Both Frame Analyzer and System Analyzer can monitor remote systems.
Thanks for the input.
1. Remote method worked. Now I am able to see GPU usage etc from another machine.
2. However, System Analyzer seems to interfere with the sample_encode process using GPU
-> When I run with-out running system Analyzer on another machine - It takes about 40 seconds to encode 1833 frames
-> When I run with system analyzer running on another machine - it takes about 230 seconds to encode 1833 frames. And I can see GPU usages initially as about 60% and going down over a period of time to about 20%. And is also reflected in encode rate, initial few 100s of frames are encoded very fast and then it slows down.
You're welcome and thanks for the performance feedback. Are these numbers all from the Intel Stick?
Do you see the same performance if you run this on a machine other than the Stick while monitoring with SA?
That is good to know that you do not see the degradation on your laptop.
As mentioned above, we do not support the Intel Sticks so I cannot say when this issue will be addressed.