Short YouTube conversational transcript comparing SVC (Scalable Video Coding) and simulcast. Simulcast is described as three independent video streams (separate RTP sync sources) at different resolutions with about 1.3× the high-resolution bitrate. SVC is a layered single bitstream where higher-resolution frames depend on lower-resolution layers and has lower overhead for similar bitrate. Practical limitations for SVC include poor hardware decoder support (notably for VP9), making simulcast the simpler and more widely adopted approach.
So how is SVC different from simulcast?
That's a good question. So let's start by describing what they are independently first. So simulcast is basically three series of frames with different resolutions and they don't depend on each other between the resolutions. So you have three different independent video streams. You have, if you go for it, 1.3 times the bitrate of the high resolution in terms of bitrate requirements. And SVC is different in that it has dependencies between the frames of different sizes. So your high resolution depends on the lower resolution.
But it's more like a single bit stream with varying layers, where each layer adds a bit more, let's say, quality or information over the previous layer.
Yes, yes. Simulcast for rtp. Simulcast is three different synchronization sources and three different transmissions and so on. While SVC is a single RTP stream.
The overhead is also smaller than simulcast for svc.
Yes, so you get more bang for the same bitrate.
Okay, so why don't you use SVC all the time?
The problem is that mainly that hardware decoders suck at it. Simulcast streams, they look like a normal video stream to a decoder. SVC needs special decoder support and in particular for VP9. That's not there yet.
So if we try to use svc, it will be a lot harder for us to find hardware encoder and decoders for them.
Yes, yes. That's a problem Google is struggling with in Google Meet, and that has been ongoing for two years, I think.
But I guess the biggest thing is that simulcast is just simpler to implement and in many cases we treat it today as good enough, at least for the last decade or so. Okay, thank you.
