Community
cancel
Showing results for 
Search instead for 
Did you mean: 
junyoung_k_
Beginner
67 Views

Windows shows 50%~100% better performance than Linux.

We tested Media SDK 2014R2 Windows Servers Vs Media SDK 2014R2 Linux Servers on same condition and hardware. 1. The result shows that Windows shows 50%~100% better performance than Linux. Let me know the reason of this result. This result comes from difference vaapi and d3d? Or API configurations?

Our test conditions as follows
- Linux : Ubuntu 12.04.5 LTS,Intel(R) Xeon(R) CPU E3-1286 v3 @ 3.70GHz,32G memory
- Windows : Microsoft Windows Server 2012 R2 Standard,Intel(R) Xeon(R) CPU E3-1286 v3 @ 3.70GHz,32G memory

2. We tested 2 hardware conditions Intel(R) Xeon(R) CPU E3-1286 v3 @ 3.70GHz Vs Intel(R) Core(TM) i7-4790 CPU @ 3.60GHz. Our test result shows that I7 hardware configuration shows 10%~30% better performance than Xeon. Let me know the reason of this result.

3. Doesn’t MediaSDK Window version support Intel(R) Core(TM) i7-4790 CPU @ 3.60GHz? Why do you support only Linux? Is there something reason for it?

0 Kudos
3 Replies
Surbhi_M_Intel
Employee
67 Views

If you are using similar conditions then you shouldn't see a difference of 50-100% increment in windows. Since the architecture of Windows and Linux is different, we can see difference in performance for long test but that should be quite small. 

  • Can you give more details of what performance evaluation you have done? like transcode/encode/decode, what is the input resolution. what were the numbers you have got? Are you using our existing samples/tutorials to evaluate? I would recommend you to use exact same conditions like using same API, choosing HW implementation, setting GPU frequency so that we are comparing apples with apples
  • Just to check are you using same OS to compare Xeon E3 and i7 4770 performance?
  • Media SDK support Intel(R) Core(TM) i7-4790 CPU @ 3.60GHz. Currently supported OS are Windows, SLES and CentOS

We have released MediaSDK 2015 R2 early this year. There are many fixes which have come in our latest releases.

Thanks,
-Surbhi

junyoung_k_
Beginner
67 Views

SURBHI M. (Intel) wrote:

If you are using similar conditions then you shouldn't see a difference of 50-100% increment in windows. Since the architecture of Windows and Linux is different, we can see difference in performance for long test but that should be quite small. 

  • Can you give more details of what performance evaluation you have done? like transcode/encode/decode, what is the input resolution. what were the numbers you have got? Are you using our existing samples/tutorials to evaluate? I would recommend you to use exact same conditions like using same API, choosing HW implementation, setting GPU frequency so that we are comparing apples with apples
  • Just to check are you using same OS to compare Xeon E3 and i7 4770 performance?
  • Media SDK support Intel(R) Core(TM) i7-4790 CPU @ 3.60GHz. Currently supported OS are Windows, SLES and CentOS

We have released MediaSDK 2015 R2 early this year. There are many fixes which have come in our latest releases.

Thanks,
-Surbhi

----------------------------------------------------------------------

1. test details

<Linux VS Window Test>

sample files : http://www.h264info.com/clips.html --> Gravity – 2K (2048×858) Trailer -> video stream(h.264)

linux:

./sample_multi_transcode_drm -i::h264 /2.h264 -o::h264 /test.out.h264 -w 1920 -h 1080 -b 5000 -deinterlace -f 30.0 -u 7 -hw
Intel(R) Media SDK Multi Transcoding Sample Version 5.0.1603344.93446

libva info: VA-API version 0.34.0
libva info: va_getDriverName() returns 0
libva info: Trying to open /usr/lib/x86_64-linux-gnu/dri/iHD_drv_video.so
libva info: Found init function __vaDriverInit_0_32
libva info: va_openDriver() returns 0
Pipeline surfaces number: min=11, max=11, chosen=11
Pipeline surfaces number: min=18, max=18, chosen=18
MFX HARDWARE Session 0 API ver 1.10 parameters:
Input  video: AVC
Output video: AVC

Session 0 was NOT joined with other sessions

Transcoding started
....................................
Transcoding finished

Common transcoding time is  16.15 sec
MFX session 0 transcoding PASSED:
Processing time: 16.15 sec
Number of processed frames: 3525

The test PASSED

windows:

D:\>sample_multi_transcode -i::h264 D:\2.h264 -o::h264 D:\test.out.h264 -w 1920 -h 1080 -b 5000 -deinterlace -f 30.0 -u 7 -hw
Intel(R) Media SDK Multi Transcoding Sample Version 5.0.461.92695

Pipeline surfaces number: min=10, max=10, chosen=10
Pipeline surfaces number: min=14, max=14, chosen=14
MFX HARDWARE Session 0 API ver 1.10 parameters:
Input  video: AVC
Output video: AVC
MFX dll: C:\Program Files\Intel\Media SDK\libmfxhw64.dll

Session 0 was NOT joined with other sessions

Transcoding started
....................................
Transcoding finished

Common transcoding time is  13.34 sec
MFX session 0 transcoding PASSED:
Processing time: 13.33 sec
Number of processed frames: 3525

The test PASSED

 

 

 

67 Views

Hi Junyoung,

Couple of things:

You are using older version of Media Server Studio. On Linux, that latest version is R3.1 and it has some critical performance bug fixes. Please use that version instead of the 1.10 version - that should close some performance gaps.

Regarding performance experiments, some suggestions - I would recommend you run long (multiple minutes) tests to evaluate the performance (or loop the files multiple times to simulate long runs). With short tests, the OS differences and how each OS handles turbo (power/turbo settings) come into play. Also, short tests do not stress the system enough, so they are not best representations. Also, from my personal experiments, Windows GPU turbo is faster than on Linux - so one reason why you "may" be seeing the difference in the execution times.

"Doesn’t MediaSDK Window version support Intel(R) Core(TM) i7-4790 CPU @ 3.60GHz? Why do you support only Linux? Is there something reason for it?" -> Can you elaborate on this please? Media Sever Studio works on both Windows and Linux today.

Reply