VAAPI Stress Test (Skylake i5-6600 Intel HD 530)

[Guest plexman]

• Introduction

So today I decided to do a simple hardware transcoding test with my setup, and share it with this amazing community that helped (and is helping) me a lot. First of all let me introduce you to my Ubuntu 16.04.2 Emby server, all my content is mp4/h264/aac-lc/web-optimized formatted for best compatibility and streaming with my aproximately 120 users (big family and a lot of university friends). So we could say that I avoid as best as I can transcoding (95% of the time the media is direct played), but sometimes users don't have a good internet connection or device and the content must be transcoded. I have hardware transcoding enabled with VAAPI as it's explained in this post. I was worried about the transcoding capabilities of the little HD 530 so that's the main reason of this test. These are my server specs:

 

CPU: Intel i5-6600

GPU: Integrated in CPU, Intel HD 530

MB: Asus B150M-A

RAM: 16 Gb DDR4-2133

OS SSD: Samsung 850 Pro SATAIII 120Gb

MEDIA HDD: 2x HGST Desktar NAS 4 Tb (8 Tb) in Software RAID 0 (I know, it's dangerous, but until I earn more money I assume the risk   . After all it's only movies, not critical data)

PSU: Corsair CX430 430W

ISP: Movistar España, 300/300 Mbps FTTH

 

Emby is running behind Apache acting as reverse proxy as seen in this post.

• The test

The test is soooo simple, I've choosen a movie in Full HD and played it at a lower quality/bitrate lots of times in Google Chrome to see what happened. The movie specs shown by Emby are the next: (BTW the movie is "8 miles"):

 

 

The choosen quality for playing was 720p - 2 mbps:

 

 

• The modest result

The result was that I was able to play the same movie Twelve (12) times in my browser, all of them transcoding (720p - 2mbps) in different processes. See the screenshots attached, one are the Chrome's Tabs and the other is HTOP showing processes and system usage (CPU and RAM).

 

I couldn't try more transcoding instances because my computer (the one where I was playing the movie) started freezing with so much chrome tabs playing videos. So I couldn't reach the real server limit.

• My personal conclusion

VAAPI is amazing (and Emby of course more amazing for including it  ), It's true that maybe quality is not as pure as software transcoding but hey! you can transcode 12+ HD videos at the same time!!! I can't imagine how much SD videos this thing can transcode, but certainly a lot. BONUS: you don't need an extra graphics card in your system that limits you to 2 transcoding instances and uses slightly more power. Now I'm in peace because I know that my server can handle the load correctly.

 

Finally, let the discussion begin!

 

PD: All transcoding logs attached.

transcoding-logs.zip

Edited July 12, 2017 by plexman

[Luke 37071]

Great info, thanks !

[Jdiesel 1114]

Good info. Would it be possible for you to run an identical test with vaapi turned off to see how software transcoding on the i5 6600 compares? Also do you think that transcoding the same source video multiple times versus transcoding multiple unique videos affects disk performance? Would be interesting to know if/when mechanical HDD performance becomes a bottleneck. 

Edited July 12, 2017 by Jdiesel

[Guest plexman]

Good info. Would it be possible for you to run an identical test with vaapi turned off to see how software transcoding on the i5 6600 compares? Also do you think that transcoding the same source video multiple times versus transcoding multiple unique videos affects disk performance? Would be interesting to know if/when mechanical HDD performance becomes a bottleneck.

I'll try to do the software transcoding test if my clients let me[emoji2] . I think that even if it's the same file, the server is reading it independently for every transcoding instance, so under hdd usage it's like reading 12 different files (IMO).

 

I think we are far of mechanical bottleneck, and further if we are talking about RAID 0 (where reading speed doubles). In my case, each HGST Desktar NAS reaches 132 Mb/s of file transfer (large files), with RAID 0 the speed is between 1.9-2 times faster. The bottleneck is my internet speed wich is limited to 37,5 Mb/s (300 Mbps / 8). So no problem for HDD [emoji6] .

 

Enviado desde mi Aquaris E5 mediante Tapatalk