Machine workload (pre-split video clips versus proxy's)

[Rogier Tuinte]

Hello,

 

We're working on a project with a widescreen projection with 4 projectors (each 1920 x 1200) 

The total resolution of the projection is about 5500px x 1200px

 

I'm wondering what is better for the workload of a Watchout machine:

 

Working with proxy's and pre-split files for each display, resulting in playing four 1920 x 1200 files?

Or positioning 3 files next to each other in 3 layers, resulting in two 1920 x 1200 files and one 1660 x 1200 file.

 

It seems to me that playing back these 3 clips is less processor intense than four 1920 x 1200 proxy's.

Am I right that Watchout is spreading the decoding of the proxy files across different cpu cores?

 

I'm asking this because at some points, where the timeline is getting complex it could happen that when i start a new background movie, one of the pre-split background clips is running out of sync with maybe 1 or 2 frames.

 

To be sure that the 3 movies are in perfect sync during playback. Does it help to put them in a composition?

Or is it better to give the clips a pre-roll with a manual value instead of the automatic pre-roll?

Does it help that I pause the timeline for a second in the middle of a pre-roll area, so that I'm sure that the new clips are fully loaded / cached?

 

And now my second and last question.

It seems to me that when I go offline and back online, clips are not starting smooth. As long as I stay online and after I played the clips for the first time, everything runs smoother.

 

When I loop clips with a jump back cue on the timeline (and jump back to a position 1 second before the clip) this runs smoother than clips with the parameter freerunning and loopable set. (question 3 ;-)) 

 

Codec:

MP4 - 50fps - about 50mbit/s - each frame is a keyframe

 

System specs:

Intel XEON E5v2 Sixcore-Processor (6x 3.5GHz, max Turbo Frequency 3.9GHz)

1x 120GB High-Performance SSD for operating system

2x 120GB High-Performance SSD for content (RAID-1)

AMD Firepro W9100, 16GB GDDR5 ECC with S400 Sync module

Watchout 5.5.2

[Walter]

Hi Rogier, can't answer your specific questions as I'm curious about the answers myself. Just a question about your encoding : 50mbps seems a lot for the resolution and codec. 25mbps should be enough in my experience, and having a higher bitrate shouldn't present any improvement in quality compared to the lower. Please share you're experience on this subject. Did you experience better quality or improvement in playback that made you choose for such a high bandwidth?

 

Vr groet

 

Walter

[screenshaper]

MP4 is just a container that can wrap a lot of codecs.

I assume you are using H.264 (AVC) video codec.

 

Having each frame as a keyframe is not a good idea.

The encoder is not allowed to do his job taking advantage of the repeating or moving parts of the pictures.

This way, you need big bitrates and get poor quality.

[Steve Farris]

I've been having struggles with .mp4 (h.264) files and have reverted to MPEG2.  I am interested in any feedback generated by this thread for clues.  One is Screenshaper's comment not to use each frame as a keyframe.  I have been doing exactly that (which could help explain the higher data rates needed).  I have found that using every frame as a keyframe helps with sync and also helps when "scrubbing" the timeline.  My understanding is the reasoning would be the same as keeping a GOP of 1 when using MPEG2.

 

If this is not the official recommendation, what should the keyframe distance be?

 

I have also wondered about playing separate videos vs using a proxy.  I don't have specific test results, but my hunch over time is that sync is better when using a proxy.  Otherwise, it seems proxies are from the days of using one screen per display machine and keeping the workload of playing other screen's videos off of machines that don't need to play them.  But now that we are regularly using multiple displays out of one machine, that consideration seems irrelevant.

[Mike Fahl]

I've frequently used a GOP structure of 15 for MPEG-2 with WATCHOUT. Make sure you use a closed GOP structure. I assume the OP is intending to run this off one machine. You may want to see if it can handle it as a single 5500px x 1200px video. I believe H.264 will allow you to make such a high resolution. MPEG-2 can be more difficult to encode beyond 1920 (although I have made several 4k wide MPEG-2 using ffmpeg as the encoder).

 

I suggest you stick to constant bit rate, as that makes the whole process simpler for the decoder. If you'äre using H.264, just use the default GOP structure suggested by your encoder. Doing it as a pre-split proxy is generally more efficient than playing multiple (possibly partially overlapping) videos on several layers, as each computer will then both play "its own" as well as any adjacent screen's videos that may partially intrude on the display.

 

Hope this helps.

 

Mike

[screenshaper]

Mike, maybe I am wrong, but I remember reccomendations not to exceed 4096 px in media width/height.

Are that specs outdated now?

[Mike Fahl]

That limitation have been increased by  factor 2 in WO6. Here we track the limitation of graphics cards in a somewhat conservative manner. So this is a hardware limitation rather than a WATCHOUT imitation. Also keep in mind that large images like this consume a lot of VRAM on the graphics card. You can determine how much space it needs using

 

   width x height x 4

 

Do the math here and make sure to stay well within your VRAM budget (keeping in mind that there are many other pieces that also use VRAM).

 

Mike