Questions on media SDK API 1.13

[david123456]

I have download media SDK API 1.13 an have some questions as follow:

1. Is it only for linux?

2. Is look ahead BRC availble for interlace?

3. What's differ between MFX_ratecontrol_CQP and MFX_ratecontrol_QVBR?

[Surbhi_M_Intel]

Hi David,

1. Is it only for linux?
 >>API 1.13 is available for both Linux and windows. 

2. Is look ahead BRC availble for interlace?
>>Yes LA BRC mode is available for interlace.

3. What's differ between MFX_ratecontrol_CQP and MFX_ratecontrol_QVBR?
>>I will get back to you for this with a detailed answer. 

Thanks,
-Surbhi

[david123456]

Thanks

1. Is it only for linux?
 >>API 1.13 is available for both Linux and windows. 

=>> OK, both for server. Does server SDK more efficency than client SDK? I see on i7 4770K for 1080p tu7 transcoding, client SDK is 180FPS and server SDK up to 300FPS.

2. Is look ahead BRC availble for interlace?
>>Yes LA BRC mode is available for interlace.

=>> OK. Is MBaff availble for H264 interlace encoding?

3. What's differ between MFX_ratecontrol_CQP and MFX_ratecontrol_QVBR?
>>I will get back to you for this with a detailed answer. 

=>> waiting the answer

[Sravanthi_K_Intel]

Hi David,

The SDK for Media has APIs that expose the underlying HW capabilities to the developer. So, with systems with better HW, the performance will be more - thus, you are observing higher perf on server than client. When you compare your server and client, make note of the system details for the both - that should give you the insights.

[david123456]

Thanks for your quick answer.

With same HW( i7 4770K), if server SDK has higher performance than client SDK? I know sever SDK need pruchase and client SDK is free.

[Surbhi_M_Intel]

Hi David,

Does server SDK more efficency than client SDK? I see on i7 4770K for 1080p tu7 transcoding, client SDK is 180FPS and server SDK up to 300FPS.
>>Media Servere Studio and MediaSDK Client both are using same API's that expose the underlying HW capabilities. Can you please confirm which OS you are using to test Server Studio? Are both results on the same OS and same driver? It is not expected to see a huge difference on same platform with same OS & driver. Media SDK client in part of a suite now - details can be found here - https://software.intel.com/en-us/intel-inde 

Is MBaff availble for H264 interlace encoding?
>>We do not have MBAFF currently for H264 encoding.  

 What's differ between MFX_ratecontrol_CQP and MFX_ratecontrol_QVBR?
>>CQP is for keeping the quantization parameter to be same for all the frame types to maintain the constant quality. 
Where as QVBR will result into constant quality but bitrate may vary dramatically. 
Perceptible image quality may seem better in QVBR because it is dynamically changing the bitrate to keep constant quality. Also, CQP encoding mode handles bit rate adjustment on a per-frame base, which is not possible with QVBR.

Thanks,
-Surbhi

[david123456]

Thanks for your clear and details answer!

I see on i7 4770K for 1080p tu7 transcoding, client SDK is 180FPS(see figure right part, it comes from http://tetrachromesoftware.com/q264Test1Analysis/q264test_3.html) and server SDK up to 300FPS (see figure left part, it comes from https://software.intel.com/en-us/intel-media-server-studio/details ). This is not my test results. I will do some tests under CentOS7 to repeat  Performance results of concurrent H.264->H.264 transcodes accelerated using Quick Sync Video using sample_multi_transcode. And then tell you my results.

 About differ between MFX_ratecontrol_CQP and MFX_ratecontrol_QVBR. I think I confused the concepts of quantization parameter and quality. Let me think it over.

About no MBAFF support in H264 encoder, I think low compression efficiency for interlace video.

[david123456]

We do some test at Centos7 i5 3790 CPU by sample_multi_transcode. The source is Crowd_run_1080p_30M.

 The transcoding rate is about 255 FPS average from 1:2~1:11 multi-tanscode. 255 FPS is lower 15% than intel's results (300FPS). I think this maybe caused by complex of Crowd_run clip and weak perf of I5 3790 comparing with I7 4770K.

[Surbhi_M_Intel]

Hi David,

I am glad you have done test yourself to measure studio server performance. 
Starting about comparison of results btw http://tetrachromesoftware.com/q264Test1Analysis/q264test_3.html and https://software.intel.com/en-us/intel-media-server-studio/details.
I don't think this is an apple to apple comparison -the set up might be quite different like the Media Server Studio released, the method of taking calculating the values, input, driver version & etc... Lot of factors to count before comparing. With every release we include lot of changes and fixes to our known issues.  

The numbers which you have got looks really good. The reason it is not 300fps or closer is because you are using a Ivy Bridge system, the method of calculating the performance. From the webpage - "Each point is an average of 4 streams and 6 bitrates with error bars showing performance variation across streams and bitrate.  Target Usage 7 (“TU7”) is the highest speed (and lowest quality) operating point.)". Also it would be good to run longer clips or repeat the test for few time for the initial warm up. 

Thanks,
-Surbhi

[david123456]

Hi, Surbhi

Thanks for your explornations.

Cerrection: The CPU type I used is I5 4590 (has. HD4600) not I5 3790. It's my error.

I use  all 4 clips to test. The results:

Crowd_run_30M     255FPS

in_to_tree              260FPS

old_town_cross     308FPS

park_joy_30M       258 FPS

If start X-window, FPS derease 10%.

By the way, I see from "intel-gfx-prm-osrc-hsw-media-vdbox.pdf" that hasDev support MBaff. If SDK will support it in near future?

[Surbhi_M_Intel]

Hi David, 

I see you rae using an i5 4590 processor, however the results posted on our website were taken on i7 4770K. you can see the difference btw those processor over here. 
Regarding Mbaff support, I have checked with the experts, we don't have this feature available in hw & sw. 
Regarding X-window decrease in performance, is because our solution is mainly focused for server where we expect the system to be headless due to which we haven't optimize our samples for the display.However there is no limitation from the SDK, you can optimize the solution according to your usecase.

Thanks,
-Surbhi

[david123456]

Hi, Surbhi

Thanks a lot.

About Mbaff, What's the meaning of following link (on page 113)?

https://01.org/linuxgraphics/sites/default/files/documentation/intel-gfx-prm-osrc-hsw-media-vdbox.pdf

For TV broadcasting application, most HDTV formats are still 1080i. This is what I care of Mbaff.