Media (Intel® oneAPI Video Processing Library, Intel Media SDK)
Access community support with transcoding, decoding, and encoding in applications using media tools from Intel. This includes Intel® oneAPI Video Processing Library and Intel® Media SDK.
This community is designed for sharing of public information. Please do not share Intel or third-party confidential information here.

QSV performance


hi ,

i find on i3 i5 platform, the QSV performance is double than soft h264 decoding by ffmpeg. but when using QSV on win7(Media SDK 2014 R2 for Clients), Only 6 channel 1080P or 8 channel 720P can decode. if add more channel, it can not decode anything. so i want to know the performance  

0 Kudos
2 Replies

Hi smart,

Thank you for the question. Currently this is not documented anywhere and is a gap in our documentation. In general, Media SDK HW decoder is very fast. The performance for the multiple decoding depends upon various factors - 

1. HW capabilities - It will vary from hardware to hardware that how many sessions can be decoded simultaneously.
2. Resolution - One should be able to do around 16 multiple decode sessions, especially at lower resolution.
3. one imp point to understand here is that more processing is required to write back the output. If you skip that, then definitely you should be able to handle more concurrent sessions. So the best method would be to do everything on the GPU as involving copying the output to the CPU can be a bottleneck.

We should be filling out this gap in the future releases.






Hi Smart,

One of the method by which you can check the no. of multiple session your system can decode is by using one of the tutorials multisession_omp present in old tutorial v0.0.2(intel_media_sdk_tutorial_041813), which can be downloaded from Media Solution Portal. With this tutorial you can run multiple decode sessions by specifying the number of tasks.
An important point is to use video memory so you should specify -d3d or -d3d11 as an input argument. Also to use -hw option, failure to mention this can run the session on the CPU which will be expensive in terms of lot of overhead copying of the data. You can measure the load of a GPU by using any tool present online. You should see change in load and GPU core clock. Also, running multiple decode session often results into effective utilization of the decoder.
Again, number of session a system can handle will differ with different systems. So there is no single number to tell the number of decode session Intel Media SDK can handle simultaneously.