I wanted to share my experience on the development of a DME prototype based on CasparCG.
You can see the result here :last test at home test in the control room
The PC configuration is a Z440 : Intel Xeon E5-1650V4 (3.2Ghz), RAM 16Go, Nvidia GeForce Titan X, with a PCIe SSD
Video Output : Decklink 4K Pro
Video Inputs : Decklink Quad 2
Everything is made with the Caspar's Mixer. To do a composition like the one in the videos above, it require the use of hundreds of layers (containing various type of producers like color, decklink, ffmpeg).
In order to achieve that I had to change a little thing in the source code of Caspar : the maximum size of the AMCP command queue wich was limited to 64 pending commands. My system send more than 1000 commands in less than a second.
There is 5 issues that I need to work on :
- The system take 10 frames of delay (even with just one layer and no mixer transformation on it)
- We can't remote it from the video switcher. I think i will work on a PBus utility
- The interlacing/deinterlacing/reintarlcing process is not very good for the quality of the product
- It doesn't work at all on the Caspar Server 2.1.0 beta
- A very weird issue : it sometimes lag and drop frames if we don't add a screen consumer in addition to the decklink consumer. With the screen consumer added, it never lag.